11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/namespace.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * (C) Copyright Al Viro 2000, 2001 51da177e4SLinus Torvalds * Released under GPL v2. 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * Based on code from fs/super.c, copyright Linus Torvalds and others. 81da177e4SLinus Torvalds * Heavily rewritten. 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds 111da177e4SLinus Torvalds #include <linux/syscalls.h> 12d10577a8SAl Viro #include <linux/export.h> 1316f7e0feSRandy Dunlap #include <linux/capability.h> 146b3286edSKirill Korotaev #include <linux/mnt_namespace.h> 15771b1371SEric W. Biederman #include <linux/user_namespace.h> 161da177e4SLinus Torvalds #include <linux/namei.h> 171da177e4SLinus Torvalds #include <linux/security.h> 185b825c3aSIngo Molnar #include <linux/cred.h> 1973cd49ecSMiklos Szeredi #include <linux/idr.h> 2057f150a5SRob Landley #include <linux/init.h> /* init_rootfs */ 21d10577a8SAl Viro #include <linux/fs_struct.h> /* get_fs_root et.al. */ 22d10577a8SAl Viro #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ 23d10577a8SAl Viro #include <linux/uaccess.h> 240bb80f24SDavid Howells #include <linux/proc_ns.h> 2520b4fb48SLinus Torvalds #include <linux/magic.h> 260818bf27SAl Viro #include <linux/bootmem.h> 279ea459e1SAl Viro #include <linux/task_work.h> 289164bb4aSIngo Molnar #include <linux/sched/task.h> 299164bb4aSIngo Molnar 3007b20889SRam Pai #include "pnode.h" 31948730b0SAdrian Bunk #include "internal.h" 321da177e4SLinus Torvalds 33d2921684SEric W. Biederman /* Maximum number of mounts in a mount namespace */ 34d2921684SEric W. Biederman unsigned int sysctl_mount_max __read_mostly = 100000; 35d2921684SEric W. Biederman 360818bf27SAl Viro static unsigned int m_hash_mask __read_mostly; 370818bf27SAl Viro static unsigned int m_hash_shift __read_mostly; 380818bf27SAl Viro static unsigned int mp_hash_mask __read_mostly; 390818bf27SAl Viro static unsigned int mp_hash_shift __read_mostly; 400818bf27SAl Viro 410818bf27SAl Viro static __initdata unsigned long mhash_entries; 420818bf27SAl Viro static int __init set_mhash_entries(char *str) 430818bf27SAl Viro { 440818bf27SAl Viro if (!str) 450818bf27SAl Viro return 0; 460818bf27SAl Viro mhash_entries = simple_strtoul(str, &str, 0); 470818bf27SAl Viro return 1; 480818bf27SAl Viro } 490818bf27SAl Viro __setup("mhash_entries=", set_mhash_entries); 500818bf27SAl Viro 510818bf27SAl Viro static __initdata unsigned long mphash_entries; 520818bf27SAl Viro static int __init set_mphash_entries(char *str) 530818bf27SAl Viro { 540818bf27SAl Viro if (!str) 550818bf27SAl Viro return 0; 560818bf27SAl Viro mphash_entries = simple_strtoul(str, &str, 0); 570818bf27SAl Viro return 1; 580818bf27SAl Viro } 590818bf27SAl Viro __setup("mphash_entries=", set_mphash_entries); 6013f14b4dSEric Dumazet 61c7999c36SAl Viro static u64 event; 6273cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 63719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 645addc5ddSAl Viro 6538129a13SAl Viro static struct hlist_head *mount_hashtable __read_mostly; 660818bf27SAl Viro static struct hlist_head *mountpoint_hashtable __read_mostly; 67e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 6859aa0da8SAl Viro static DECLARE_RWSEM(namespace_sem); 691da177e4SLinus Torvalds 70f87fd4c2SMiklos Szeredi /* /sys/fs */ 7100d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 7200d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 73f87fd4c2SMiklos Szeredi 7499b7db7bSNick Piggin /* 7599b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 7699b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 7799b7db7bSNick Piggin * up the tree. 7899b7db7bSNick Piggin * 7999b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 8099b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 8199b7db7bSNick Piggin */ 8248a066e7SAl Viro __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); 8399b7db7bSNick Piggin 8438129a13SAl Viro static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) 851da177e4SLinus Torvalds { 861da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 871da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 880818bf27SAl Viro tmp = tmp + (tmp >> m_hash_shift); 890818bf27SAl Viro return &mount_hashtable[tmp & m_hash_mask]; 900818bf27SAl Viro } 910818bf27SAl Viro 920818bf27SAl Viro static inline struct hlist_head *mp_hash(struct dentry *dentry) 930818bf27SAl Viro { 940818bf27SAl Viro unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); 950818bf27SAl Viro tmp = tmp + (tmp >> mp_hash_shift); 960818bf27SAl Viro return &mountpoint_hashtable[tmp & mp_hash_mask]; 971da177e4SLinus Torvalds } 981da177e4SLinus Torvalds 99b105e270SAl Viro static int mnt_alloc_id(struct mount *mnt) 10073cd49ecSMiklos Szeredi { 101169b480eSMatthew Wilcox int res = ida_alloc(&mnt_id_ida, GFP_KERNEL); 10273cd49ecSMiklos Szeredi 103169b480eSMatthew Wilcox if (res < 0) 10473cd49ecSMiklos Szeredi return res; 105169b480eSMatthew Wilcox mnt->mnt_id = res; 106169b480eSMatthew Wilcox return 0; 10773cd49ecSMiklos Szeredi } 10873cd49ecSMiklos Szeredi 109b105e270SAl Viro static void mnt_free_id(struct mount *mnt) 11073cd49ecSMiklos Szeredi { 111169b480eSMatthew Wilcox ida_free(&mnt_id_ida, mnt->mnt_id); 11273cd49ecSMiklos Szeredi } 11373cd49ecSMiklos Szeredi 114719f5d7fSMiklos Szeredi /* 115719f5d7fSMiklos Szeredi * Allocate a new peer group ID 116719f5d7fSMiklos Szeredi */ 1174b8b21f4SAl Viro static int mnt_alloc_group_id(struct mount *mnt) 118719f5d7fSMiklos Szeredi { 119169b480eSMatthew Wilcox int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL); 120f21f6220SAl Viro 121169b480eSMatthew Wilcox if (res < 0) 122f21f6220SAl Viro return res; 123169b480eSMatthew Wilcox mnt->mnt_group_id = res; 124169b480eSMatthew Wilcox return 0; 125719f5d7fSMiklos Szeredi } 126719f5d7fSMiklos Szeredi 127719f5d7fSMiklos Szeredi /* 128719f5d7fSMiklos Szeredi * Release a peer group ID 129719f5d7fSMiklos Szeredi */ 1304b8b21f4SAl Viro void mnt_release_group_id(struct mount *mnt) 131719f5d7fSMiklos Szeredi { 132169b480eSMatthew Wilcox ida_free(&mnt_group_ida, mnt->mnt_group_id); 13315169fe7SAl Viro mnt->mnt_group_id = 0; 134719f5d7fSMiklos Szeredi } 135719f5d7fSMiklos Szeredi 136b3e19d92SNick Piggin /* 137b3e19d92SNick Piggin * vfsmount lock must be held for read 138b3e19d92SNick Piggin */ 13983adc753SAl Viro static inline void mnt_add_count(struct mount *mnt, int n) 140b3e19d92SNick Piggin { 141b3e19d92SNick Piggin #ifdef CONFIG_SMP 14268e8a9feSAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, n); 143b3e19d92SNick Piggin #else 144b3e19d92SNick Piggin preempt_disable(); 14568e8a9feSAl Viro mnt->mnt_count += n; 146b3e19d92SNick Piggin preempt_enable(); 147b3e19d92SNick Piggin #endif 148b3e19d92SNick Piggin } 149b3e19d92SNick Piggin 150b3e19d92SNick Piggin /* 151b3e19d92SNick Piggin * vfsmount lock must be held for write 152b3e19d92SNick Piggin */ 15383adc753SAl Viro unsigned int mnt_get_count(struct mount *mnt) 154b3e19d92SNick Piggin { 155b3e19d92SNick Piggin #ifdef CONFIG_SMP 156f03c6599SAl Viro unsigned int count = 0; 157b3e19d92SNick Piggin int cpu; 158b3e19d92SNick Piggin 159b3e19d92SNick Piggin for_each_possible_cpu(cpu) { 16068e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; 161b3e19d92SNick Piggin } 162b3e19d92SNick Piggin 163b3e19d92SNick Piggin return count; 164b3e19d92SNick Piggin #else 16568e8a9feSAl Viro return mnt->mnt_count; 166b3e19d92SNick Piggin #endif 167b3e19d92SNick Piggin } 168b3e19d92SNick Piggin 16987b95ce0SAl Viro static void drop_mountpoint(struct fs_pin *p) 17087b95ce0SAl Viro { 17187b95ce0SAl Viro struct mount *m = container_of(p, struct mount, mnt_umount); 17287b95ce0SAl Viro dput(m->mnt_ex_mountpoint); 17387b95ce0SAl Viro pin_remove(p); 17487b95ce0SAl Viro mntput(&m->mnt); 17587b95ce0SAl Viro } 17687b95ce0SAl Viro 177b105e270SAl Viro static struct mount *alloc_vfsmnt(const char *name) 1781da177e4SLinus Torvalds { 179c63181e6SAl Viro struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 180c63181e6SAl Viro if (mnt) { 18173cd49ecSMiklos Szeredi int err; 18273cd49ecSMiklos Szeredi 183c63181e6SAl Viro err = mnt_alloc_id(mnt); 18488b38782SLi Zefan if (err) 18588b38782SLi Zefan goto out_free_cache; 18688b38782SLi Zefan 18788b38782SLi Zefan if (name) { 188fcc139aeSAndrzej Hajda mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); 189c63181e6SAl Viro if (!mnt->mnt_devname) 19088b38782SLi Zefan goto out_free_id; 19173cd49ecSMiklos Szeredi } 19273cd49ecSMiklos Szeredi 193b3e19d92SNick Piggin #ifdef CONFIG_SMP 194c63181e6SAl Viro mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 195c63181e6SAl Viro if (!mnt->mnt_pcp) 196b3e19d92SNick Piggin goto out_free_devname; 197b3e19d92SNick Piggin 198c63181e6SAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, 1); 199b3e19d92SNick Piggin #else 200c63181e6SAl Viro mnt->mnt_count = 1; 201c63181e6SAl Viro mnt->mnt_writers = 0; 202b3e19d92SNick Piggin #endif 203b3e19d92SNick Piggin 20438129a13SAl Viro INIT_HLIST_NODE(&mnt->mnt_hash); 205c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_child); 206c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_mounts); 207c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_list); 208c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_expire); 209c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_share); 210c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave_list); 211c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave); 2120a5eb7c8SEric W. Biederman INIT_HLIST_NODE(&mnt->mnt_mp_list); 21399b19d16SEric W. Biederman INIT_LIST_HEAD(&mnt->mnt_umounting); 21487b95ce0SAl Viro init_fs_pin(&mnt->mnt_umount, drop_mountpoint); 2151da177e4SLinus Torvalds } 216c63181e6SAl Viro return mnt; 21788b38782SLi Zefan 218d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 219d3ef3d73Snpiggin@suse.de out_free_devname: 220fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 221d3ef3d73Snpiggin@suse.de #endif 22288b38782SLi Zefan out_free_id: 223c63181e6SAl Viro mnt_free_id(mnt); 22488b38782SLi Zefan out_free_cache: 225c63181e6SAl Viro kmem_cache_free(mnt_cache, mnt); 22688b38782SLi Zefan return NULL; 2271da177e4SLinus Torvalds } 2281da177e4SLinus Torvalds 2298366025eSDave Hansen /* 2308366025eSDave Hansen * Most r/o checks on a fs are for operations that take 2318366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 2328366025eSDave Hansen * We must keep track of when those operations start 2338366025eSDave Hansen * (for permission checks) and when they end, so that 2348366025eSDave Hansen * we can determine when writes are able to occur to 2358366025eSDave Hansen * a filesystem. 2368366025eSDave Hansen */ 2373d733633SDave Hansen /* 2383d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 2393d733633SDave Hansen * @mnt: the mount to check for its write status 2403d733633SDave Hansen * 2413d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 2423d733633SDave Hansen * It does not guarantee that the filesystem will stay 2433d733633SDave Hansen * r/w, just that it is right *now*. This can not and 2443d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 2453d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 2463d733633SDave Hansen * r/w. 2473d733633SDave Hansen */ 2483d733633SDave Hansen int __mnt_is_readonly(struct vfsmount *mnt) 2493d733633SDave Hansen { 2502e4b7fcdSDave Hansen if (mnt->mnt_flags & MNT_READONLY) 2512e4b7fcdSDave Hansen return 1; 252bc98a42cSDavid Howells if (sb_rdonly(mnt->mnt_sb)) 2532e4b7fcdSDave Hansen return 1; 2542e4b7fcdSDave Hansen return 0; 2553d733633SDave Hansen } 2563d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2573d733633SDave Hansen 25883adc753SAl Viro static inline void mnt_inc_writers(struct mount *mnt) 2593d733633SDave Hansen { 260d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 26168e8a9feSAl Viro this_cpu_inc(mnt->mnt_pcp->mnt_writers); 262d3ef3d73Snpiggin@suse.de #else 26368e8a9feSAl Viro mnt->mnt_writers++; 264d3ef3d73Snpiggin@suse.de #endif 2653d733633SDave Hansen } 2663d733633SDave Hansen 26783adc753SAl Viro static inline void mnt_dec_writers(struct mount *mnt) 2683d733633SDave Hansen { 269d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 27068e8a9feSAl Viro this_cpu_dec(mnt->mnt_pcp->mnt_writers); 271d3ef3d73Snpiggin@suse.de #else 27268e8a9feSAl Viro mnt->mnt_writers--; 273d3ef3d73Snpiggin@suse.de #endif 274d3ef3d73Snpiggin@suse.de } 275d3ef3d73Snpiggin@suse.de 27683adc753SAl Viro static unsigned int mnt_get_writers(struct mount *mnt) 277d3ef3d73Snpiggin@suse.de { 278d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 279d3ef3d73Snpiggin@suse.de unsigned int count = 0; 2803d733633SDave Hansen int cpu; 2813d733633SDave Hansen 2823d733633SDave Hansen for_each_possible_cpu(cpu) { 28368e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 2843d733633SDave Hansen } 2853d733633SDave Hansen 286d3ef3d73Snpiggin@suse.de return count; 287d3ef3d73Snpiggin@suse.de #else 288d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 289d3ef3d73Snpiggin@suse.de #endif 2903d733633SDave Hansen } 2913d733633SDave Hansen 2924ed5e82fSMiklos Szeredi static int mnt_is_readonly(struct vfsmount *mnt) 2934ed5e82fSMiklos Szeredi { 2944ed5e82fSMiklos Szeredi if (mnt->mnt_sb->s_readonly_remount) 2954ed5e82fSMiklos Szeredi return 1; 2964ed5e82fSMiklos Szeredi /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ 2974ed5e82fSMiklos Szeredi smp_rmb(); 2984ed5e82fSMiklos Szeredi return __mnt_is_readonly(mnt); 2994ed5e82fSMiklos Szeredi } 3004ed5e82fSMiklos Szeredi 3013d733633SDave Hansen /* 302eb04c282SJan Kara * Most r/o & frozen checks on a fs are for operations that take discrete 303eb04c282SJan Kara * amounts of time, like a write() or unlink(). We must keep track of when 304eb04c282SJan Kara * those operations start (for permission checks) and when they end, so that we 305eb04c282SJan Kara * can determine when writes are able to occur to a filesystem. 3063d733633SDave Hansen */ 3078366025eSDave Hansen /** 308eb04c282SJan Kara * __mnt_want_write - get write access to a mount without freeze protection 30983adc753SAl Viro * @m: the mount on which to take a write 3108366025eSDave Hansen * 311eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 312eb04c282SJan Kara * it, and makes sure that writes are allowed (mnt it read-write) before 313eb04c282SJan Kara * returning success. This operation does not protect against filesystem being 314eb04c282SJan Kara * frozen. When the write operation is finished, __mnt_drop_write() must be 315eb04c282SJan Kara * called. This is effectively a refcount. 3168366025eSDave Hansen */ 317eb04c282SJan Kara int __mnt_want_write(struct vfsmount *m) 3188366025eSDave Hansen { 31983adc753SAl Viro struct mount *mnt = real_mount(m); 3203d733633SDave Hansen int ret = 0; 3213d733633SDave Hansen 322d3ef3d73Snpiggin@suse.de preempt_disable(); 323c6653a83SNick Piggin mnt_inc_writers(mnt); 324d3ef3d73Snpiggin@suse.de /* 325c6653a83SNick Piggin * The store to mnt_inc_writers must be visible before we pass 326d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 327d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 328d3ef3d73Snpiggin@suse.de */ 329d3ef3d73Snpiggin@suse.de smp_mb(); 3306aa7de05SMark Rutland while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) 331d3ef3d73Snpiggin@suse.de cpu_relax(); 332d3ef3d73Snpiggin@suse.de /* 333d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 334d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 335d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 336d3ef3d73Snpiggin@suse.de */ 337d3ef3d73Snpiggin@suse.de smp_rmb(); 3384ed5e82fSMiklos Szeredi if (mnt_is_readonly(m)) { 339c6653a83SNick Piggin mnt_dec_writers(mnt); 3403d733633SDave Hansen ret = -EROFS; 3413d733633SDave Hansen } 342d3ef3d73Snpiggin@suse.de preempt_enable(); 343eb04c282SJan Kara 344eb04c282SJan Kara return ret; 345eb04c282SJan Kara } 346eb04c282SJan Kara 347eb04c282SJan Kara /** 348eb04c282SJan Kara * mnt_want_write - get write access to a mount 349eb04c282SJan Kara * @m: the mount on which to take a write 350eb04c282SJan Kara * 351eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 352eb04c282SJan Kara * it, and makes sure that writes are allowed (mount is read-write, filesystem 353eb04c282SJan Kara * is not frozen) before returning success. When the write operation is 354eb04c282SJan Kara * finished, mnt_drop_write() must be called. This is effectively a refcount. 355eb04c282SJan Kara */ 356eb04c282SJan Kara int mnt_want_write(struct vfsmount *m) 357eb04c282SJan Kara { 358eb04c282SJan Kara int ret; 359eb04c282SJan Kara 360eb04c282SJan Kara sb_start_write(m->mnt_sb); 361eb04c282SJan Kara ret = __mnt_want_write(m); 362eb04c282SJan Kara if (ret) 363eb04c282SJan Kara sb_end_write(m->mnt_sb); 3643d733633SDave Hansen return ret; 3658366025eSDave Hansen } 3668366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 3678366025eSDave Hansen 3688366025eSDave Hansen /** 36996029c4eSnpiggin@suse.de * mnt_clone_write - get write access to a mount 37096029c4eSnpiggin@suse.de * @mnt: the mount on which to take a write 37196029c4eSnpiggin@suse.de * 37296029c4eSnpiggin@suse.de * This is effectively like mnt_want_write, except 37396029c4eSnpiggin@suse.de * it must only be used to take an extra write reference 37496029c4eSnpiggin@suse.de * on a mountpoint that we already know has a write reference 37596029c4eSnpiggin@suse.de * on it. This allows some optimisation. 37696029c4eSnpiggin@suse.de * 37796029c4eSnpiggin@suse.de * After finished, mnt_drop_write must be called as usual to 37896029c4eSnpiggin@suse.de * drop the reference. 37996029c4eSnpiggin@suse.de */ 38096029c4eSnpiggin@suse.de int mnt_clone_write(struct vfsmount *mnt) 38196029c4eSnpiggin@suse.de { 38296029c4eSnpiggin@suse.de /* superblock may be r/o */ 38396029c4eSnpiggin@suse.de if (__mnt_is_readonly(mnt)) 38496029c4eSnpiggin@suse.de return -EROFS; 38596029c4eSnpiggin@suse.de preempt_disable(); 38683adc753SAl Viro mnt_inc_writers(real_mount(mnt)); 38796029c4eSnpiggin@suse.de preempt_enable(); 38896029c4eSnpiggin@suse.de return 0; 38996029c4eSnpiggin@suse.de } 39096029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_clone_write); 39196029c4eSnpiggin@suse.de 39296029c4eSnpiggin@suse.de /** 393eb04c282SJan Kara * __mnt_want_write_file - get write access to a file's mount 394eb04c282SJan Kara * @file: the file who's mount on which to take a write 395eb04c282SJan Kara * 396eb04c282SJan Kara * This is like __mnt_want_write, but it takes a file and can 397eb04c282SJan Kara * do some optimisations if the file is open for write already 398eb04c282SJan Kara */ 399eb04c282SJan Kara int __mnt_want_write_file(struct file *file) 400eb04c282SJan Kara { 40183f936c7SAl Viro if (!(file->f_mode & FMODE_WRITER)) 402eb04c282SJan Kara return __mnt_want_write(file->f_path.mnt); 403eb04c282SJan Kara else 404eb04c282SJan Kara return mnt_clone_write(file->f_path.mnt); 405eb04c282SJan Kara } 406eb04c282SJan Kara 407eb04c282SJan Kara /** 4087c6893e3SMiklos Szeredi * mnt_want_write_file_path - get write access to a file's mount 40996029c4eSnpiggin@suse.de * @file: the file who's mount on which to take a write 41096029c4eSnpiggin@suse.de * 41196029c4eSnpiggin@suse.de * This is like mnt_want_write, but it takes a file and can 41296029c4eSnpiggin@suse.de * do some optimisations if the file is open for write already 4137c6893e3SMiklos Szeredi * 4147c6893e3SMiklos Szeredi * Called by the vfs for cases when we have an open file at hand, but will do an 4157c6893e3SMiklos Szeredi * inode operation on it (important distinction for files opened on overlayfs, 4167c6893e3SMiklos Szeredi * since the file operations will come from the real underlying file, while 4177c6893e3SMiklos Szeredi * inode operations come from the overlay). 41896029c4eSnpiggin@suse.de */ 4197c6893e3SMiklos Szeredi int mnt_want_write_file_path(struct file *file) 42096029c4eSnpiggin@suse.de { 421eb04c282SJan Kara int ret; 422eb04c282SJan Kara 423eb04c282SJan Kara sb_start_write(file->f_path.mnt->mnt_sb); 424eb04c282SJan Kara ret = __mnt_want_write_file(file); 425eb04c282SJan Kara if (ret) 426eb04c282SJan Kara sb_end_write(file->f_path.mnt->mnt_sb); 427eb04c282SJan Kara return ret; 42896029c4eSnpiggin@suse.de } 4297c6893e3SMiklos Szeredi 4307c6893e3SMiklos Szeredi static inline int may_write_real(struct file *file) 4317c6893e3SMiklos Szeredi { 4327c6893e3SMiklos Szeredi struct dentry *dentry = file->f_path.dentry; 4337c6893e3SMiklos Szeredi struct dentry *upperdentry; 4347c6893e3SMiklos Szeredi 4357c6893e3SMiklos Szeredi /* Writable file? */ 4367c6893e3SMiklos Szeredi if (file->f_mode & FMODE_WRITER) 4377c6893e3SMiklos Szeredi return 0; 4387c6893e3SMiklos Szeredi 4397c6893e3SMiklos Szeredi /* Not overlayfs? */ 4407c6893e3SMiklos Szeredi if (likely(!(dentry->d_flags & DCACHE_OP_REAL))) 4417c6893e3SMiklos Szeredi return 0; 4427c6893e3SMiklos Szeredi 4437c6893e3SMiklos Szeredi /* File refers to upper, writable layer? */ 4447c6893e3SMiklos Szeredi upperdentry = d_real(dentry, NULL, 0, D_REAL_UPPER); 445954c736fSAmir Goldstein if (upperdentry && 446954c736fSAmir Goldstein (file_inode(file) == d_inode(upperdentry) || 447954c736fSAmir Goldstein file_inode(file) == d_inode(dentry))) 4487c6893e3SMiklos Szeredi return 0; 4497c6893e3SMiklos Szeredi 4507c6893e3SMiklos Szeredi /* Lower layer: can't write to real file, sorry... */ 4517c6893e3SMiklos Szeredi return -EPERM; 4527c6893e3SMiklos Szeredi } 4537c6893e3SMiklos Szeredi 4547c6893e3SMiklos Szeredi /** 4557c6893e3SMiklos Szeredi * mnt_want_write_file - get write access to a file's mount 4567c6893e3SMiklos Szeredi * @file: the file who's mount on which to take a write 4577c6893e3SMiklos Szeredi * 4587c6893e3SMiklos Szeredi * This is like mnt_want_write, but it takes a file and can 4597c6893e3SMiklos Szeredi * do some optimisations if the file is open for write already 4607c6893e3SMiklos Szeredi * 4617c6893e3SMiklos Szeredi * Mostly called by filesystems from their ioctl operation before performing 4627c6893e3SMiklos Szeredi * modification. On overlayfs this needs to check if the file is on a read-only 4637c6893e3SMiklos Szeredi * lower layer and deny access in that case. 4647c6893e3SMiklos Szeredi */ 4657c6893e3SMiklos Szeredi int mnt_want_write_file(struct file *file) 4667c6893e3SMiklos Szeredi { 4677c6893e3SMiklos Szeredi int ret; 4687c6893e3SMiklos Szeredi 4697c6893e3SMiklos Szeredi ret = may_write_real(file); 4707c6893e3SMiklos Szeredi if (!ret) { 4717c6893e3SMiklos Szeredi sb_start_write(file_inode(file)->i_sb); 4727c6893e3SMiklos Szeredi ret = __mnt_want_write_file(file); 4737c6893e3SMiklos Szeredi if (ret) 4747c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 4757c6893e3SMiklos Szeredi } 4767c6893e3SMiklos Szeredi return ret; 4777c6893e3SMiklos Szeredi } 47896029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 47996029c4eSnpiggin@suse.de 48096029c4eSnpiggin@suse.de /** 481eb04c282SJan Kara * __mnt_drop_write - give up write access to a mount 4828366025eSDave Hansen * @mnt: the mount on which to give up write access 4838366025eSDave Hansen * 4848366025eSDave Hansen * Tells the low-level filesystem that we are done 4858366025eSDave Hansen * performing writes to it. Must be matched with 486eb04c282SJan Kara * __mnt_want_write() call above. 4878366025eSDave Hansen */ 488eb04c282SJan Kara void __mnt_drop_write(struct vfsmount *mnt) 4898366025eSDave Hansen { 490d3ef3d73Snpiggin@suse.de preempt_disable(); 49183adc753SAl Viro mnt_dec_writers(real_mount(mnt)); 492d3ef3d73Snpiggin@suse.de preempt_enable(); 4938366025eSDave Hansen } 494eb04c282SJan Kara 495eb04c282SJan Kara /** 496eb04c282SJan Kara * mnt_drop_write - give up write access to a mount 497eb04c282SJan Kara * @mnt: the mount on which to give up write access 498eb04c282SJan Kara * 499eb04c282SJan Kara * Tells the low-level filesystem that we are done performing writes to it and 500eb04c282SJan Kara * also allows filesystem to be frozen again. Must be matched with 501eb04c282SJan Kara * mnt_want_write() call above. 502eb04c282SJan Kara */ 503eb04c282SJan Kara void mnt_drop_write(struct vfsmount *mnt) 504eb04c282SJan Kara { 505eb04c282SJan Kara __mnt_drop_write(mnt); 506eb04c282SJan Kara sb_end_write(mnt->mnt_sb); 507eb04c282SJan Kara } 5088366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 5098366025eSDave Hansen 510eb04c282SJan Kara void __mnt_drop_write_file(struct file *file) 511eb04c282SJan Kara { 512eb04c282SJan Kara __mnt_drop_write(file->f_path.mnt); 513eb04c282SJan Kara } 514eb04c282SJan Kara 5157c6893e3SMiklos Szeredi void mnt_drop_write_file_path(struct file *file) 5162a79f17eSAl Viro { 5172a79f17eSAl Viro mnt_drop_write(file->f_path.mnt); 5182a79f17eSAl Viro } 5197c6893e3SMiklos Szeredi 5207c6893e3SMiklos Szeredi void mnt_drop_write_file(struct file *file) 5217c6893e3SMiklos Szeredi { 5227c6893e3SMiklos Szeredi __mnt_drop_write(file->f_path.mnt); 5237c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 5247c6893e3SMiklos Szeredi } 5252a79f17eSAl Viro EXPORT_SYMBOL(mnt_drop_write_file); 5262a79f17eSAl Viro 52783adc753SAl Viro static int mnt_make_readonly(struct mount *mnt) 5288366025eSDave Hansen { 5293d733633SDave Hansen int ret = 0; 5303d733633SDave Hansen 531719ea2fbSAl Viro lock_mount_hash(); 53283adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 533d3ef3d73Snpiggin@suse.de /* 534d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 535d3ef3d73Snpiggin@suse.de * should be visible before we do. 536d3ef3d73Snpiggin@suse.de */ 537d3ef3d73Snpiggin@suse.de smp_mb(); 538d3ef3d73Snpiggin@suse.de 539d3ef3d73Snpiggin@suse.de /* 540d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 541d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 542d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 543d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 544d3ef3d73Snpiggin@suse.de * 545d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 546d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 547d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 548d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 549d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 550d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 551d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 552d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 553d3ef3d73Snpiggin@suse.de * we're counting up here. 554d3ef3d73Snpiggin@suse.de */ 555c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 556d3ef3d73Snpiggin@suse.de ret = -EBUSY; 557d3ef3d73Snpiggin@suse.de else 55883adc753SAl Viro mnt->mnt.mnt_flags |= MNT_READONLY; 559d3ef3d73Snpiggin@suse.de /* 560d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 561d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 562d3ef3d73Snpiggin@suse.de */ 563d3ef3d73Snpiggin@suse.de smp_wmb(); 56483adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 565719ea2fbSAl Viro unlock_mount_hash(); 5663d733633SDave Hansen return ret; 5673d733633SDave Hansen } 5688366025eSDave Hansen 56983adc753SAl Viro static void __mnt_unmake_readonly(struct mount *mnt) 5702e4b7fcdSDave Hansen { 571719ea2fbSAl Viro lock_mount_hash(); 57283adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_READONLY; 573719ea2fbSAl Viro unlock_mount_hash(); 5742e4b7fcdSDave Hansen } 5752e4b7fcdSDave Hansen 5764ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 5774ed5e82fSMiklos Szeredi { 5784ed5e82fSMiklos Szeredi struct mount *mnt; 5794ed5e82fSMiklos Szeredi int err = 0; 5804ed5e82fSMiklos Szeredi 5818e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 5828e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 5838e8b8796SMiklos Szeredi return -EBUSY; 5848e8b8796SMiklos Szeredi 585719ea2fbSAl Viro lock_mount_hash(); 5864ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5874ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 5884ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 5894ed5e82fSMiklos Szeredi smp_mb(); 5904ed5e82fSMiklos Szeredi if (mnt_get_writers(mnt) > 0) { 5914ed5e82fSMiklos Szeredi err = -EBUSY; 5924ed5e82fSMiklos Szeredi break; 5934ed5e82fSMiklos Szeredi } 5944ed5e82fSMiklos Szeredi } 5954ed5e82fSMiklos Szeredi } 5968e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 5978e8b8796SMiklos Szeredi err = -EBUSY; 5988e8b8796SMiklos Szeredi 5994ed5e82fSMiklos Szeredi if (!err) { 6004ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 6014ed5e82fSMiklos Szeredi smp_wmb(); 6024ed5e82fSMiklos Szeredi } 6034ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 6044ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 6054ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 6064ed5e82fSMiklos Szeredi } 607719ea2fbSAl Viro unlock_mount_hash(); 6084ed5e82fSMiklos Szeredi 6094ed5e82fSMiklos Szeredi return err; 6104ed5e82fSMiklos Szeredi } 6114ed5e82fSMiklos Szeredi 612b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 6131da177e4SLinus Torvalds { 614fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 615d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 61668e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 617d3ef3d73Snpiggin@suse.de #endif 618b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 6191da177e4SLinus Torvalds } 6201da177e4SLinus Torvalds 6218ffcb32eSDavid Howells static void delayed_free_vfsmnt(struct rcu_head *head) 6228ffcb32eSDavid Howells { 6238ffcb32eSDavid Howells free_vfsmnt(container_of(head, struct mount, mnt_rcu)); 6248ffcb32eSDavid Howells } 6258ffcb32eSDavid Howells 62648a066e7SAl Viro /* call under rcu_read_lock */ 627294d71ffSAl Viro int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) 62848a066e7SAl Viro { 62948a066e7SAl Viro struct mount *mnt; 63048a066e7SAl Viro if (read_seqretry(&mount_lock, seq)) 631294d71ffSAl Viro return 1; 63248a066e7SAl Viro if (bastard == NULL) 633294d71ffSAl Viro return 0; 63448a066e7SAl Viro mnt = real_mount(bastard); 63548a066e7SAl Viro mnt_add_count(mnt, 1); 636119e1ef8SAl Viro smp_mb(); // see mntput_no_expire() 63748a066e7SAl Viro if (likely(!read_seqretry(&mount_lock, seq))) 638294d71ffSAl Viro return 0; 63948a066e7SAl Viro if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { 64048a066e7SAl Viro mnt_add_count(mnt, -1); 641294d71ffSAl Viro return 1; 64248a066e7SAl Viro } 643119e1ef8SAl Viro lock_mount_hash(); 644119e1ef8SAl Viro if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { 645119e1ef8SAl Viro mnt_add_count(mnt, -1); 646119e1ef8SAl Viro unlock_mount_hash(); 647119e1ef8SAl Viro return 1; 648119e1ef8SAl Viro } 649119e1ef8SAl Viro unlock_mount_hash(); 650119e1ef8SAl Viro /* caller will mntput() */ 651294d71ffSAl Viro return -1; 652294d71ffSAl Viro } 653294d71ffSAl Viro 654294d71ffSAl Viro /* call under rcu_read_lock */ 655294d71ffSAl Viro bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) 656294d71ffSAl Viro { 657294d71ffSAl Viro int res = __legitimize_mnt(bastard, seq); 658294d71ffSAl Viro if (likely(!res)) 659294d71ffSAl Viro return true; 660294d71ffSAl Viro if (unlikely(res < 0)) { 66148a066e7SAl Viro rcu_read_unlock(); 66248a066e7SAl Viro mntput(bastard); 66348a066e7SAl Viro rcu_read_lock(); 664294d71ffSAl Viro } 66548a066e7SAl Viro return false; 66648a066e7SAl Viro } 66748a066e7SAl Viro 6681da177e4SLinus Torvalds /* 669474279dcSAl Viro * find the first mount at @dentry on vfsmount @mnt. 67048a066e7SAl Viro * call under rcu_read_lock() 6711da177e4SLinus Torvalds */ 672474279dcSAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 6731da177e4SLinus Torvalds { 67438129a13SAl Viro struct hlist_head *head = m_hash(mnt, dentry); 675474279dcSAl Viro struct mount *p; 6761da177e4SLinus Torvalds 67738129a13SAl Viro hlist_for_each_entry_rcu(p, head, mnt_hash) 678474279dcSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) 679474279dcSAl Viro return p; 680474279dcSAl Viro return NULL; 6811da177e4SLinus Torvalds } 682474279dcSAl Viro 683474279dcSAl Viro /* 684f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 685f015f126SDavid Howells * 686f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 687f015f126SDavid Howells * following mounts: 688f015f126SDavid Howells * 689f015f126SDavid Howells * mount /dev/sda1 /mnt 690f015f126SDavid Howells * mount /dev/sda2 /mnt 691f015f126SDavid Howells * mount /dev/sda3 /mnt 692f015f126SDavid Howells * 693f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 694f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 695f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 696f015f126SDavid Howells * 697f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 698a05964f3SRam Pai */ 699ca71cf71SAl Viro struct vfsmount *lookup_mnt(const struct path *path) 700a05964f3SRam Pai { 701c7105365SAl Viro struct mount *child_mnt; 70248a066e7SAl Viro struct vfsmount *m; 70348a066e7SAl Viro unsigned seq; 70499b7db7bSNick Piggin 70548a066e7SAl Viro rcu_read_lock(); 70648a066e7SAl Viro do { 70748a066e7SAl Viro seq = read_seqbegin(&mount_lock); 708474279dcSAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry); 70948a066e7SAl Viro m = child_mnt ? &child_mnt->mnt : NULL; 71048a066e7SAl Viro } while (!legitimize_mnt(m, seq)); 71148a066e7SAl Viro rcu_read_unlock(); 71248a066e7SAl Viro return m; 713a05964f3SRam Pai } 714a05964f3SRam Pai 7157af1364fSEric W. Biederman /* 7167af1364fSEric W. Biederman * __is_local_mountpoint - Test to see if dentry is a mountpoint in the 7177af1364fSEric W. Biederman * current mount namespace. 7187af1364fSEric W. Biederman * 7197af1364fSEric W. Biederman * The common case is dentries are not mountpoints at all and that 7207af1364fSEric W. Biederman * test is handled inline. For the slow case when we are actually 7217af1364fSEric W. Biederman * dealing with a mountpoint of some kind, walk through all of the 7227af1364fSEric W. Biederman * mounts in the current mount namespace and test to see if the dentry 7237af1364fSEric W. Biederman * is a mountpoint. 7247af1364fSEric W. Biederman * 7257af1364fSEric W. Biederman * The mount_hashtable is not usable in the context because we 7267af1364fSEric W. Biederman * need to identify all mounts that may be in the current mount 7277af1364fSEric W. Biederman * namespace not just a mount that happens to have some specified 7287af1364fSEric W. Biederman * parent mount. 7297af1364fSEric W. Biederman */ 7307af1364fSEric W. Biederman bool __is_local_mountpoint(struct dentry *dentry) 7317af1364fSEric W. Biederman { 7327af1364fSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 7337af1364fSEric W. Biederman struct mount *mnt; 7347af1364fSEric W. Biederman bool is_covered = false; 7357af1364fSEric W. Biederman 7367af1364fSEric W. Biederman if (!d_mountpoint(dentry)) 7377af1364fSEric W. Biederman goto out; 7387af1364fSEric W. Biederman 7397af1364fSEric W. Biederman down_read(&namespace_sem); 7407af1364fSEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 7417af1364fSEric W. Biederman is_covered = (mnt->mnt_mountpoint == dentry); 7427af1364fSEric W. Biederman if (is_covered) 7437af1364fSEric W. Biederman break; 7447af1364fSEric W. Biederman } 7457af1364fSEric W. Biederman up_read(&namespace_sem); 7467af1364fSEric W. Biederman out: 7477af1364fSEric W. Biederman return is_covered; 7487af1364fSEric W. Biederman } 7497af1364fSEric W. Biederman 750e2dfa935SEric W. Biederman static struct mountpoint *lookup_mountpoint(struct dentry *dentry) 75184d17192SAl Viro { 7520818bf27SAl Viro struct hlist_head *chain = mp_hash(dentry); 75384d17192SAl Viro struct mountpoint *mp; 75484d17192SAl Viro 7550818bf27SAl Viro hlist_for_each_entry(mp, chain, m_hash) { 75684d17192SAl Viro if (mp->m_dentry == dentry) { 75784d17192SAl Viro /* might be worth a WARN_ON() */ 75884d17192SAl Viro if (d_unlinked(dentry)) 75984d17192SAl Viro return ERR_PTR(-ENOENT); 76084d17192SAl Viro mp->m_count++; 76184d17192SAl Viro return mp; 76284d17192SAl Viro } 76384d17192SAl Viro } 764e2dfa935SEric W. Biederman return NULL; 765e2dfa935SEric W. Biederman } 766e2dfa935SEric W. Biederman 7673895dbf8SEric W. Biederman static struct mountpoint *get_mountpoint(struct dentry *dentry) 768e2dfa935SEric W. Biederman { 7693895dbf8SEric W. Biederman struct mountpoint *mp, *new = NULL; 770e2dfa935SEric W. Biederman int ret; 77184d17192SAl Viro 7723895dbf8SEric W. Biederman if (d_mountpoint(dentry)) { 7733895dbf8SEric W. Biederman mountpoint: 7743895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7753895dbf8SEric W. Biederman mp = lookup_mountpoint(dentry); 7763895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7773895dbf8SEric W. Biederman if (mp) 7783895dbf8SEric W. Biederman goto done; 77984d17192SAl Viro } 780eed81007SMiklos Szeredi 7813895dbf8SEric W. Biederman if (!new) 7823895dbf8SEric W. Biederman new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 7833895dbf8SEric W. Biederman if (!new) 7843895dbf8SEric W. Biederman return ERR_PTR(-ENOMEM); 7853895dbf8SEric W. Biederman 7863895dbf8SEric W. Biederman 7873895dbf8SEric W. Biederman /* Exactly one processes may set d_mounted */ 7883895dbf8SEric W. Biederman ret = d_set_mounted(dentry); 7893895dbf8SEric W. Biederman 7903895dbf8SEric W. Biederman /* Someone else set d_mounted? */ 7913895dbf8SEric W. Biederman if (ret == -EBUSY) 7923895dbf8SEric W. Biederman goto mountpoint; 7933895dbf8SEric W. Biederman 7943895dbf8SEric W. Biederman /* The dentry is not available as a mountpoint? */ 7953895dbf8SEric W. Biederman mp = ERR_PTR(ret); 7963895dbf8SEric W. Biederman if (ret) 7973895dbf8SEric W. Biederman goto done; 7983895dbf8SEric W. Biederman 7993895dbf8SEric W. Biederman /* Add the new mountpoint to the hash table */ 8003895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 8013895dbf8SEric W. Biederman new->m_dentry = dentry; 8023895dbf8SEric W. Biederman new->m_count = 1; 8033895dbf8SEric W. Biederman hlist_add_head(&new->m_hash, mp_hash(dentry)); 8043895dbf8SEric W. Biederman INIT_HLIST_HEAD(&new->m_list); 8053895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 8063895dbf8SEric W. Biederman 8073895dbf8SEric W. Biederman mp = new; 8083895dbf8SEric W. Biederman new = NULL; 8093895dbf8SEric W. Biederman done: 8103895dbf8SEric W. Biederman kfree(new); 81184d17192SAl Viro return mp; 81284d17192SAl Viro } 81384d17192SAl Viro 81484d17192SAl Viro static void put_mountpoint(struct mountpoint *mp) 81584d17192SAl Viro { 81684d17192SAl Viro if (!--mp->m_count) { 81784d17192SAl Viro struct dentry *dentry = mp->m_dentry; 8180a5eb7c8SEric W. Biederman BUG_ON(!hlist_empty(&mp->m_list)); 81984d17192SAl Viro spin_lock(&dentry->d_lock); 82084d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 82184d17192SAl Viro spin_unlock(&dentry->d_lock); 8220818bf27SAl Viro hlist_del(&mp->m_hash); 82384d17192SAl Viro kfree(mp); 82484d17192SAl Viro } 82584d17192SAl Viro } 82684d17192SAl Viro 827143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 8281da177e4SLinus Torvalds { 8296b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 8301da177e4SLinus Torvalds } 8311da177e4SLinus Torvalds 83299b7db7bSNick Piggin /* 83399b7db7bSNick Piggin * vfsmount lock must be held for write 83499b7db7bSNick Piggin */ 8356b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 8365addc5ddSAl Viro { 8375addc5ddSAl Viro if (ns) { 8385addc5ddSAl Viro ns->event = ++event; 8395addc5ddSAl Viro wake_up_interruptible(&ns->poll); 8405addc5ddSAl Viro } 8415addc5ddSAl Viro } 8425addc5ddSAl Viro 84399b7db7bSNick Piggin /* 84499b7db7bSNick Piggin * vfsmount lock must be held for write 84599b7db7bSNick Piggin */ 8466b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 8475addc5ddSAl Viro { 8485addc5ddSAl Viro if (ns && ns->event != event) { 8495addc5ddSAl Viro ns->event = event; 8505addc5ddSAl Viro wake_up_interruptible(&ns->poll); 8515addc5ddSAl Viro } 8525addc5ddSAl Viro } 8535addc5ddSAl Viro 85499b7db7bSNick Piggin /* 85599b7db7bSNick Piggin * vfsmount lock must be held for write 85699b7db7bSNick Piggin */ 8577bdb11deSEric W. Biederman static void unhash_mnt(struct mount *mnt) 8581da177e4SLinus Torvalds { 8590714a533SAl Viro mnt->mnt_parent = mnt; 860a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8616b41d536SAl Viro list_del_init(&mnt->mnt_child); 86238129a13SAl Viro hlist_del_init_rcu(&mnt->mnt_hash); 8630a5eb7c8SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 86484d17192SAl Viro put_mountpoint(mnt->mnt_mp); 86584d17192SAl Viro mnt->mnt_mp = NULL; 8661da177e4SLinus Torvalds } 8671da177e4SLinus Torvalds 86899b7db7bSNick Piggin /* 86999b7db7bSNick Piggin * vfsmount lock must be held for write 87099b7db7bSNick Piggin */ 8717bdb11deSEric W. Biederman static void detach_mnt(struct mount *mnt, struct path *old_path) 8727bdb11deSEric W. Biederman { 8737bdb11deSEric W. Biederman old_path->dentry = mnt->mnt_mountpoint; 8747bdb11deSEric W. Biederman old_path->mnt = &mnt->mnt_parent->mnt; 8757bdb11deSEric W. Biederman unhash_mnt(mnt); 8767bdb11deSEric W. Biederman } 8777bdb11deSEric W. Biederman 8787bdb11deSEric W. Biederman /* 8797bdb11deSEric W. Biederman * vfsmount lock must be held for write 8807bdb11deSEric W. Biederman */ 8816a46c573SEric W. Biederman static void umount_mnt(struct mount *mnt) 8826a46c573SEric W. Biederman { 8836a46c573SEric W. Biederman /* old mountpoint will be dropped when we can do that */ 8846a46c573SEric W. Biederman mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint; 8856a46c573SEric W. Biederman unhash_mnt(mnt); 8866a46c573SEric W. Biederman } 8876a46c573SEric W. Biederman 8886a46c573SEric W. Biederman /* 8896a46c573SEric W. Biederman * vfsmount lock must be held for write 8906a46c573SEric W. Biederman */ 89184d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 89284d17192SAl Viro struct mountpoint *mp, 89344d964d6SAl Viro struct mount *child_mnt) 894b90fa9aeSRam Pai { 89584d17192SAl Viro mp->m_count++; 8963a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 89784d17192SAl Viro child_mnt->mnt_mountpoint = dget(mp->m_dentry); 8983a2393d7SAl Viro child_mnt->mnt_parent = mnt; 89984d17192SAl Viro child_mnt->mnt_mp = mp; 9000a5eb7c8SEric W. Biederman hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); 901b90fa9aeSRam Pai } 902b90fa9aeSRam Pai 9031064f874SEric W. Biederman static void __attach_mnt(struct mount *mnt, struct mount *parent) 9041064f874SEric W. Biederman { 9051064f874SEric W. Biederman hlist_add_head_rcu(&mnt->mnt_hash, 9061064f874SEric W. Biederman m_hash(&parent->mnt, mnt->mnt_mountpoint)); 9071064f874SEric W. Biederman list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 9081064f874SEric W. Biederman } 9091064f874SEric W. Biederman 91099b7db7bSNick Piggin /* 91199b7db7bSNick Piggin * vfsmount lock must be held for write 91299b7db7bSNick Piggin */ 91384d17192SAl Viro static void attach_mnt(struct mount *mnt, 91484d17192SAl Viro struct mount *parent, 91584d17192SAl Viro struct mountpoint *mp) 9161da177e4SLinus Torvalds { 91784d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 9181064f874SEric W. Biederman __attach_mnt(mnt, parent); 919b90fa9aeSRam Pai } 920b90fa9aeSRam Pai 9211064f874SEric W. Biederman void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) 92212a5b529SAl Viro { 9231064f874SEric W. Biederman struct mountpoint *old_mp = mnt->mnt_mp; 9241064f874SEric W. Biederman struct dentry *old_mountpoint = mnt->mnt_mountpoint; 9251064f874SEric W. Biederman struct mount *old_parent = mnt->mnt_parent; 9261064f874SEric W. Biederman 9271064f874SEric W. Biederman list_del_init(&mnt->mnt_child); 9281064f874SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 9291064f874SEric W. Biederman hlist_del_init_rcu(&mnt->mnt_hash); 9301064f874SEric W. Biederman 9311064f874SEric W. Biederman attach_mnt(mnt, parent, mp); 9321064f874SEric W. Biederman 9331064f874SEric W. Biederman put_mountpoint(old_mp); 9341064f874SEric W. Biederman 9351064f874SEric W. Biederman /* 9361064f874SEric W. Biederman * Safely avoid even the suggestion this code might sleep or 9371064f874SEric W. Biederman * lock the mount hash by taking advantage of the knowledge that 9381064f874SEric W. Biederman * mnt_change_mountpoint will not release the final reference 9391064f874SEric W. Biederman * to a mountpoint. 9401064f874SEric W. Biederman * 9411064f874SEric W. Biederman * During mounting, the mount passed in as the parent mount will 9421064f874SEric W. Biederman * continue to use the old mountpoint and during unmounting, the 9431064f874SEric W. Biederman * old mountpoint will continue to exist until namespace_unlock, 9441064f874SEric W. Biederman * which happens well after mnt_change_mountpoint. 9451064f874SEric W. Biederman */ 9461064f874SEric W. Biederman spin_lock(&old_mountpoint->d_lock); 9471064f874SEric W. Biederman old_mountpoint->d_lockref.count--; 9481064f874SEric W. Biederman spin_unlock(&old_mountpoint->d_lock); 9491064f874SEric W. Biederman 9501064f874SEric W. Biederman mnt_add_count(old_parent, -1); 95112a5b529SAl Viro } 95212a5b529SAl Viro 953b90fa9aeSRam Pai /* 95499b7db7bSNick Piggin * vfsmount lock must be held for write 955b90fa9aeSRam Pai */ 9561064f874SEric W. Biederman static void commit_tree(struct mount *mnt) 957b90fa9aeSRam Pai { 9580714a533SAl Viro struct mount *parent = mnt->mnt_parent; 95983adc753SAl Viro struct mount *m; 960b90fa9aeSRam Pai LIST_HEAD(head); 961143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 962b90fa9aeSRam Pai 9630714a533SAl Viro BUG_ON(parent == mnt); 964b90fa9aeSRam Pai 9651a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 966f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 967143c8c91SAl Viro m->mnt_ns = n; 968f03c6599SAl Viro 969b90fa9aeSRam Pai list_splice(&head, n->list.prev); 970b90fa9aeSRam Pai 971d2921684SEric W. Biederman n->mounts += n->pending_mounts; 972d2921684SEric W. Biederman n->pending_mounts = 0; 973d2921684SEric W. Biederman 9741064f874SEric W. Biederman __attach_mnt(mnt, parent); 9756b3286edSKirill Korotaev touch_mnt_namespace(n); 9761da177e4SLinus Torvalds } 9771da177e4SLinus Torvalds 978909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 9791da177e4SLinus Torvalds { 9806b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 9816b41d536SAl Viro if (next == &p->mnt_mounts) { 9821da177e4SLinus Torvalds while (1) { 983909b0a88SAl Viro if (p == root) 9841da177e4SLinus Torvalds return NULL; 9856b41d536SAl Viro next = p->mnt_child.next; 9866b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 9871da177e4SLinus Torvalds break; 9880714a533SAl Viro p = p->mnt_parent; 9891da177e4SLinus Torvalds } 9901da177e4SLinus Torvalds } 9916b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 9921da177e4SLinus Torvalds } 9931da177e4SLinus Torvalds 994315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 9959676f0c6SRam Pai { 9966b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 9976b41d536SAl Viro while (prev != &p->mnt_mounts) { 9986b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 9996b41d536SAl Viro prev = p->mnt_mounts.prev; 10009676f0c6SRam Pai } 10019676f0c6SRam Pai return p; 10029676f0c6SRam Pai } 10039676f0c6SRam Pai 10049d412a43SAl Viro struct vfsmount * 10059d412a43SAl Viro vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) 10069d412a43SAl Viro { 1007b105e270SAl Viro struct mount *mnt; 10089d412a43SAl Viro struct dentry *root; 10099d412a43SAl Viro 10109d412a43SAl Viro if (!type) 10119d412a43SAl Viro return ERR_PTR(-ENODEV); 10129d412a43SAl Viro 10139d412a43SAl Viro mnt = alloc_vfsmnt(name); 10149d412a43SAl Viro if (!mnt) 10159d412a43SAl Viro return ERR_PTR(-ENOMEM); 10169d412a43SAl Viro 1017e462ec50SDavid Howells if (flags & SB_KERNMOUNT) 1018b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 10199d412a43SAl Viro 10209d412a43SAl Viro root = mount_fs(type, flags, name, data); 10219d412a43SAl Viro if (IS_ERR(root)) { 10228ffcb32eSDavid Howells mnt_free_id(mnt); 10239d412a43SAl Viro free_vfsmnt(mnt); 10249d412a43SAl Viro return ERR_CAST(root); 10259d412a43SAl Viro } 10269d412a43SAl Viro 1027b105e270SAl Viro mnt->mnt.mnt_root = root; 1028b105e270SAl Viro mnt->mnt.mnt_sb = root->d_sb; 1029a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 10300714a533SAl Viro mnt->mnt_parent = mnt; 1031719ea2fbSAl Viro lock_mount_hash(); 103239f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); 1033719ea2fbSAl Viro unlock_mount_hash(); 1034b105e270SAl Viro return &mnt->mnt; 10359d412a43SAl Viro } 10369d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 10379d412a43SAl Viro 103893faccbbSEric W. Biederman struct vfsmount * 103993faccbbSEric W. Biederman vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, 104093faccbbSEric W. Biederman const char *name, void *data) 104193faccbbSEric W. Biederman { 104293faccbbSEric W. Biederman /* Until it is worked out how to pass the user namespace 104393faccbbSEric W. Biederman * through from the parent mount to the submount don't support 104493faccbbSEric W. Biederman * unprivileged mounts with submounts. 104593faccbbSEric W. Biederman */ 104693faccbbSEric W. Biederman if (mountpoint->d_sb->s_user_ns != &init_user_ns) 104793faccbbSEric W. Biederman return ERR_PTR(-EPERM); 104893faccbbSEric W. Biederman 1049e462ec50SDavid Howells return vfs_kern_mount(type, SB_SUBMOUNT, name, data); 105093faccbbSEric W. Biederman } 105193faccbbSEric W. Biederman EXPORT_SYMBOL_GPL(vfs_submount); 105293faccbbSEric W. Biederman 105387129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 105436341f64SRam Pai int flag) 10551da177e4SLinus Torvalds { 105687129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 1057be34d1a3SDavid Howells struct mount *mnt; 1058be34d1a3SDavid Howells int err; 10591da177e4SLinus Torvalds 1060be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 1061be34d1a3SDavid Howells if (!mnt) 1062be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 1063be34d1a3SDavid Howells 10647a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 106515169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 1066719f5d7fSMiklos Szeredi else 106715169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 1068719f5d7fSMiklos Szeredi 106915169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 1070be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 1071719f5d7fSMiklos Szeredi if (err) 1072719f5d7fSMiklos Szeredi goto out_free; 1073719f5d7fSMiklos Szeredi } 1074719f5d7fSMiklos Szeredi 107516a34adbSAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags; 107616a34adbSAl Viro mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); 1077132c94e3SEric W. Biederman /* Don't allow unprivileged users to change mount flags */ 10789566d674SEric W. Biederman if (flag & CL_UNPRIVILEGED) { 10799566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_ATIME; 10809566d674SEric W. Biederman 10819566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_READONLY) 1082132c94e3SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; 1083132c94e3SEric W. Biederman 10849566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NODEV) 10859566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NODEV; 10869566d674SEric W. Biederman 10879566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NOSUID) 10889566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID; 10899566d674SEric W. Biederman 10909566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NOEXEC) 10919566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC; 10929566d674SEric W. Biederman } 10939566d674SEric W. Biederman 10945ff9d8a6SEric W. Biederman /* Don't allow unprivileged users to reveal what is under a mount */ 1095381cacb1SEric W. Biederman if ((flag & CL_UNPRIVILEGED) && 1096381cacb1SEric W. Biederman (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire))) 10975ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCKED; 10985ff9d8a6SEric W. Biederman 10991da177e4SLinus Torvalds atomic_inc(&sb->s_active); 1100b105e270SAl Viro mnt->mnt.mnt_sb = sb; 1101b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 1102a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 11030714a533SAl Viro mnt->mnt_parent = mnt; 1104719ea2fbSAl Viro lock_mount_hash(); 110539f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 1106719ea2fbSAl Viro unlock_mount_hash(); 1107b90fa9aeSRam Pai 11087a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 11097a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 11106776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 111132301920SAl Viro mnt->mnt_master = old; 1112fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 11138aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 1114fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 11156776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 1116d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 11176776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 1118d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 11195235d448SAl Viro } else { 11205235d448SAl Viro CLEAR_MNT_SHARED(mnt); 11215afe0022SRam Pai } 1122b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 11230f0afb1dSAl Viro set_mnt_shared(mnt); 11241da177e4SLinus Torvalds 11251da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 11261da177e4SLinus Torvalds * as the original if that was on one */ 112736341f64SRam Pai if (flag & CL_EXPIRE) { 11286776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 11296776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 11301da177e4SLinus Torvalds } 1131be34d1a3SDavid Howells 1132cb338d06SAl Viro return mnt; 1133719f5d7fSMiklos Szeredi 1134719f5d7fSMiklos Szeredi out_free: 11358ffcb32eSDavid Howells mnt_free_id(mnt); 1136719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 1137be34d1a3SDavid Howells return ERR_PTR(err); 11381da177e4SLinus Torvalds } 11391da177e4SLinus Torvalds 11409ea459e1SAl Viro static void cleanup_mnt(struct mount *mnt) 11419ea459e1SAl Viro { 11429ea459e1SAl Viro /* 11439ea459e1SAl Viro * This probably indicates that somebody messed 11449ea459e1SAl Viro * up a mnt_want/drop_write() pair. If this 11459ea459e1SAl Viro * happens, the filesystem was probably unable 11469ea459e1SAl Viro * to make r/w->r/o transitions. 11479ea459e1SAl Viro */ 11489ea459e1SAl Viro /* 11499ea459e1SAl Viro * The locking used to deal with mnt_count decrement provides barriers, 11509ea459e1SAl Viro * so mnt_get_writers() below is safe. 11519ea459e1SAl Viro */ 11529ea459e1SAl Viro WARN_ON(mnt_get_writers(mnt)); 11539ea459e1SAl Viro if (unlikely(mnt->mnt_pins.first)) 11549ea459e1SAl Viro mnt_pin_kill(mnt); 11559ea459e1SAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 11569ea459e1SAl Viro dput(mnt->mnt.mnt_root); 11579ea459e1SAl Viro deactivate_super(mnt->mnt.mnt_sb); 11589ea459e1SAl Viro mnt_free_id(mnt); 11599ea459e1SAl Viro call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); 11609ea459e1SAl Viro } 11619ea459e1SAl Viro 11629ea459e1SAl Viro static void __cleanup_mnt(struct rcu_head *head) 11639ea459e1SAl Viro { 11649ea459e1SAl Viro cleanup_mnt(container_of(head, struct mount, mnt_rcu)); 11659ea459e1SAl Viro } 11669ea459e1SAl Viro 11679ea459e1SAl Viro static LLIST_HEAD(delayed_mntput_list); 11689ea459e1SAl Viro static void delayed_mntput(struct work_struct *unused) 11699ea459e1SAl Viro { 11709ea459e1SAl Viro struct llist_node *node = llist_del_all(&delayed_mntput_list); 117129785735SByungchul Park struct mount *m, *t; 11729ea459e1SAl Viro 117329785735SByungchul Park llist_for_each_entry_safe(m, t, node, mnt_llist) 117429785735SByungchul Park cleanup_mnt(m); 11759ea459e1SAl Viro } 11769ea459e1SAl Viro static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); 11779ea459e1SAl Viro 1178900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 11797b7b1aceSAl Viro { 118048a066e7SAl Viro rcu_read_lock(); 11819ea0a46cSAl Viro if (likely(READ_ONCE(mnt->mnt_ns))) { 11829ea0a46cSAl Viro /* 11839ea0a46cSAl Viro * Since we don't do lock_mount_hash() here, 11849ea0a46cSAl Viro * ->mnt_ns can change under us. However, if it's 11859ea0a46cSAl Viro * non-NULL, then there's a reference that won't 11869ea0a46cSAl Viro * be dropped until after an RCU delay done after 11879ea0a46cSAl Viro * turning ->mnt_ns NULL. So if we observe it 11889ea0a46cSAl Viro * non-NULL under rcu_read_lock(), the reference 11899ea0a46cSAl Viro * we are dropping is not the final one. 11909ea0a46cSAl Viro */ 1191aa9c0e07SAl Viro mnt_add_count(mnt, -1); 119248a066e7SAl Viro rcu_read_unlock(); 119399b7db7bSNick Piggin return; 1194b3e19d92SNick Piggin } 1195719ea2fbSAl Viro lock_mount_hash(); 1196119e1ef8SAl Viro /* 1197119e1ef8SAl Viro * make sure that if __legitimize_mnt() has not seen us grab 1198119e1ef8SAl Viro * mount_lock, we'll see their refcount increment here. 1199119e1ef8SAl Viro */ 1200119e1ef8SAl Viro smp_mb(); 12019ea0a46cSAl Viro mnt_add_count(mnt, -1); 1202b3e19d92SNick Piggin if (mnt_get_count(mnt)) { 120348a066e7SAl Viro rcu_read_unlock(); 1204719ea2fbSAl Viro unlock_mount_hash(); 120599b7db7bSNick Piggin return; 120699b7db7bSNick Piggin } 120748a066e7SAl Viro if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { 120848a066e7SAl Viro rcu_read_unlock(); 120948a066e7SAl Viro unlock_mount_hash(); 121048a066e7SAl Viro return; 121148a066e7SAl Viro } 121248a066e7SAl Viro mnt->mnt.mnt_flags |= MNT_DOOMED; 121348a066e7SAl Viro rcu_read_unlock(); 1214962830dfSAndi Kleen 121539f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 1216ce07d891SEric W. Biederman 1217ce07d891SEric W. Biederman if (unlikely(!list_empty(&mnt->mnt_mounts))) { 1218ce07d891SEric W. Biederman struct mount *p, *tmp; 1219ce07d891SEric W. Biederman list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { 1220ce07d891SEric W. Biederman umount_mnt(p); 1221ce07d891SEric W. Biederman } 1222ce07d891SEric W. Biederman } 1223719ea2fbSAl Viro unlock_mount_hash(); 1224649a795aSAl Viro 12259ea459e1SAl Viro if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { 12269ea459e1SAl Viro struct task_struct *task = current; 12279ea459e1SAl Viro if (likely(!(task->flags & PF_KTHREAD))) { 12289ea459e1SAl Viro init_task_work(&mnt->mnt_rcu, __cleanup_mnt); 12299ea459e1SAl Viro if (!task_work_add(task, &mnt->mnt_rcu, true)) 12309ea459e1SAl Viro return; 12319ea459e1SAl Viro } 12329ea459e1SAl Viro if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) 12339ea459e1SAl Viro schedule_delayed_work(&delayed_mntput_work, 1); 12349ea459e1SAl Viro return; 12359ea459e1SAl Viro } 12369ea459e1SAl Viro cleanup_mnt(mnt); 1237b3e19d92SNick Piggin } 1238b3e19d92SNick Piggin 1239b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 1240b3e19d92SNick Piggin { 1241b3e19d92SNick Piggin if (mnt) { 1242863d684fSAl Viro struct mount *m = real_mount(mnt); 1243b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 1244863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 1245863d684fSAl Viro m->mnt_expiry_mark = 0; 1246863d684fSAl Viro mntput_no_expire(m); 1247b3e19d92SNick Piggin } 1248b3e19d92SNick Piggin } 1249b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 1250b3e19d92SNick Piggin 1251b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 1252b3e19d92SNick Piggin { 1253b3e19d92SNick Piggin if (mnt) 125483adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 1255b3e19d92SNick Piggin return mnt; 1256b3e19d92SNick Piggin } 1257b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 1258b3e19d92SNick Piggin 1259c6609c0aSIan Kent /* path_is_mountpoint() - Check if path is a mount in the current 1260c6609c0aSIan Kent * namespace. 1261c6609c0aSIan Kent * 1262c6609c0aSIan Kent * d_mountpoint() can only be used reliably to establish if a dentry is 1263c6609c0aSIan Kent * not mounted in any namespace and that common case is handled inline. 1264c6609c0aSIan Kent * d_mountpoint() isn't aware of the possibility there may be multiple 1265c6609c0aSIan Kent * mounts using a given dentry in a different namespace. This function 1266c6609c0aSIan Kent * checks if the passed in path is a mountpoint rather than the dentry 1267c6609c0aSIan Kent * alone. 1268c6609c0aSIan Kent */ 1269c6609c0aSIan Kent bool path_is_mountpoint(const struct path *path) 1270c6609c0aSIan Kent { 1271c6609c0aSIan Kent unsigned seq; 1272c6609c0aSIan Kent bool res; 1273c6609c0aSIan Kent 1274c6609c0aSIan Kent if (!d_mountpoint(path->dentry)) 1275c6609c0aSIan Kent return false; 1276c6609c0aSIan Kent 1277c6609c0aSIan Kent rcu_read_lock(); 1278c6609c0aSIan Kent do { 1279c6609c0aSIan Kent seq = read_seqbegin(&mount_lock); 1280c6609c0aSIan Kent res = __path_is_mountpoint(path); 1281c6609c0aSIan Kent } while (read_seqretry(&mount_lock, seq)); 1282c6609c0aSIan Kent rcu_read_unlock(); 1283c6609c0aSIan Kent 1284c6609c0aSIan Kent return res; 1285c6609c0aSIan Kent } 1286c6609c0aSIan Kent EXPORT_SYMBOL(path_is_mountpoint); 1287c6609c0aSIan Kent 1288ca71cf71SAl Viro struct vfsmount *mnt_clone_internal(const struct path *path) 12897b7b1aceSAl Viro { 12903064c356SAl Viro struct mount *p; 12913064c356SAl Viro p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); 12923064c356SAl Viro if (IS_ERR(p)) 12933064c356SAl Viro return ERR_CAST(p); 12943064c356SAl Viro p->mnt.mnt_flags |= MNT_INTERNAL; 12953064c356SAl Viro return &p->mnt; 12967b7b1aceSAl Viro } 12971da177e4SLinus Torvalds 1298a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 12990226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 13001da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 13011da177e4SLinus Torvalds { 1302ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13031da177e4SLinus Torvalds 1304390c6843SRam Pai down_read(&namespace_sem); 1305c7999c36SAl Viro if (p->cached_event == p->ns->event) { 1306c7999c36SAl Viro void *v = p->cached_mount; 1307c7999c36SAl Viro if (*pos == p->cached_index) 1308c7999c36SAl Viro return v; 1309c7999c36SAl Viro if (*pos == p->cached_index + 1) { 1310c7999c36SAl Viro v = seq_list_next(v, &p->ns->list, &p->cached_index); 1311c7999c36SAl Viro return p->cached_mount = v; 1312c7999c36SAl Viro } 1313c7999c36SAl Viro } 1314c7999c36SAl Viro 1315c7999c36SAl Viro p->cached_event = p->ns->event; 1316c7999c36SAl Viro p->cached_mount = seq_list_start(&p->ns->list, *pos); 1317c7999c36SAl Viro p->cached_index = *pos; 1318c7999c36SAl Viro return p->cached_mount; 13191da177e4SLinus Torvalds } 13201da177e4SLinus Torvalds 13211da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 13221da177e4SLinus Torvalds { 1323ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 1324b0765fb8SPavel Emelianov 1325c7999c36SAl Viro p->cached_mount = seq_list_next(v, &p->ns->list, pos); 1326c7999c36SAl Viro p->cached_index = *pos; 1327c7999c36SAl Viro return p->cached_mount; 13281da177e4SLinus Torvalds } 13291da177e4SLinus Torvalds 13301da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 13311da177e4SLinus Torvalds { 1332390c6843SRam Pai up_read(&namespace_sem); 13331da177e4SLinus Torvalds } 13341da177e4SLinus Torvalds 13350226f492SAl Viro static int m_show(struct seq_file *m, void *v) 13369f5596afSAl Viro { 1337ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13381a4eeaf2SAl Viro struct mount *r = list_entry(v, struct mount, mnt_list); 13390226f492SAl Viro return p->show(m, &r->mnt); 13401da177e4SLinus Torvalds } 13411da177e4SLinus Torvalds 1342a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 13431da177e4SLinus Torvalds .start = m_start, 13441da177e4SLinus Torvalds .next = m_next, 13451da177e4SLinus Torvalds .stop = m_stop, 13460226f492SAl Viro .show = m_show, 1347b4629fe2SChuck Lever }; 1348a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1349b4629fe2SChuck Lever 13501da177e4SLinus Torvalds /** 13511da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 13521da177e4SLinus Torvalds * @mnt: root of mount tree 13531da177e4SLinus Torvalds * 13541da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 13551da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 13561da177e4SLinus Torvalds * busy. 13571da177e4SLinus Torvalds */ 1358909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 13591da177e4SLinus Torvalds { 1360909b0a88SAl Viro struct mount *mnt = real_mount(m); 136136341f64SRam Pai int actual_refs = 0; 136236341f64SRam Pai int minimum_refs = 0; 1363315fc83eSAl Viro struct mount *p; 1364909b0a88SAl Viro BUG_ON(!m); 13651da177e4SLinus Torvalds 1366b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1367719ea2fbSAl Viro lock_mount_hash(); 1368909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 136983adc753SAl Viro actual_refs += mnt_get_count(p); 13701da177e4SLinus Torvalds minimum_refs += 2; 13711da177e4SLinus Torvalds } 1372719ea2fbSAl Viro unlock_mount_hash(); 13731da177e4SLinus Torvalds 13741da177e4SLinus Torvalds if (actual_refs > minimum_refs) 13751da177e4SLinus Torvalds return 0; 1376e3474a8eSIan Kent 1377e3474a8eSIan Kent return 1; 13781da177e4SLinus Torvalds } 13791da177e4SLinus Torvalds 13801da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 13811da177e4SLinus Torvalds 13821da177e4SLinus Torvalds /** 13831da177e4SLinus Torvalds * may_umount - check if a mount point is busy 13841da177e4SLinus Torvalds * @mnt: root of mount 13851da177e4SLinus Torvalds * 13861da177e4SLinus Torvalds * This is called to check if a mount point has any 13871da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 13881da177e4SLinus Torvalds * mount has sub mounts this will return busy 13891da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 13901da177e4SLinus Torvalds * 13911da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 13921da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 13931da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 13941da177e4SLinus Torvalds */ 13951da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 13961da177e4SLinus Torvalds { 1397e3474a8eSIan Kent int ret = 1; 13988ad08d8aSAl Viro down_read(&namespace_sem); 1399719ea2fbSAl Viro lock_mount_hash(); 14001ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1401e3474a8eSIan Kent ret = 0; 1402719ea2fbSAl Viro unlock_mount_hash(); 14038ad08d8aSAl Viro up_read(&namespace_sem); 1404a05964f3SRam Pai return ret; 14051da177e4SLinus Torvalds } 14061da177e4SLinus Torvalds 14071da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 14081da177e4SLinus Torvalds 140938129a13SAl Viro static HLIST_HEAD(unmounted); /* protected by namespace_sem */ 1410e3197d83SAl Viro 141197216be0SAl Viro static void namespace_unlock(void) 14121da177e4SLinus Torvalds { 1413a3b3c562SEric W. Biederman struct hlist_head head; 141497216be0SAl Viro 1415a3b3c562SEric W. Biederman hlist_move_list(&unmounted, &head); 1416a3b3c562SEric W. Biederman 141797216be0SAl Viro up_write(&namespace_sem); 1418a3b3c562SEric W. Biederman 1419a3b3c562SEric W. Biederman if (likely(hlist_empty(&head))) 142097216be0SAl Viro return; 142197216be0SAl Viro 142248a066e7SAl Viro synchronize_rcu(); 142348a066e7SAl Viro 142487b95ce0SAl Viro group_pin_kill(&head); 142570fbcdf4SRam Pai } 142670fbcdf4SRam Pai 142797216be0SAl Viro static inline void namespace_lock(void) 1428e3197d83SAl Viro { 142997216be0SAl Viro down_write(&namespace_sem); 1430e3197d83SAl Viro } 1431e3197d83SAl Viro 1432e819f152SEric W. Biederman enum umount_tree_flags { 1433e819f152SEric W. Biederman UMOUNT_SYNC = 1, 1434e819f152SEric W. Biederman UMOUNT_PROPAGATE = 2, 1435e0c9c0afSEric W. Biederman UMOUNT_CONNECTED = 4, 1436e819f152SEric W. Biederman }; 1437f2d0a123SEric W. Biederman 1438f2d0a123SEric W. Biederman static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) 1439f2d0a123SEric W. Biederman { 1440f2d0a123SEric W. Biederman /* Leaving mounts connected is only valid for lazy umounts */ 1441f2d0a123SEric W. Biederman if (how & UMOUNT_SYNC) 1442f2d0a123SEric W. Biederman return true; 1443f2d0a123SEric W. Biederman 1444f2d0a123SEric W. Biederman /* A mount without a parent has nothing to be connected to */ 1445f2d0a123SEric W. Biederman if (!mnt_has_parent(mnt)) 1446f2d0a123SEric W. Biederman return true; 1447f2d0a123SEric W. Biederman 1448f2d0a123SEric W. Biederman /* Because the reference counting rules change when mounts are 1449f2d0a123SEric W. Biederman * unmounted and connected, umounted mounts may not be 1450f2d0a123SEric W. Biederman * connected to mounted mounts. 1451f2d0a123SEric W. Biederman */ 1452f2d0a123SEric W. Biederman if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) 1453f2d0a123SEric W. Biederman return true; 1454f2d0a123SEric W. Biederman 1455f2d0a123SEric W. Biederman /* Has it been requested that the mount remain connected? */ 1456f2d0a123SEric W. Biederman if (how & UMOUNT_CONNECTED) 1457f2d0a123SEric W. Biederman return false; 1458f2d0a123SEric W. Biederman 1459f2d0a123SEric W. Biederman /* Is the mount locked such that it needs to remain connected? */ 1460f2d0a123SEric W. Biederman if (IS_MNT_LOCKED(mnt)) 1461f2d0a123SEric W. Biederman return false; 1462f2d0a123SEric W. Biederman 1463f2d0a123SEric W. Biederman /* By default disconnect the mount */ 1464f2d0a123SEric W. Biederman return true; 1465f2d0a123SEric W. Biederman } 1466f2d0a123SEric W. Biederman 146799b7db7bSNick Piggin /* 146848a066e7SAl Viro * mount_lock must be held 146999b7db7bSNick Piggin * namespace_sem must be held for write 147099b7db7bSNick Piggin */ 1471e819f152SEric W. Biederman static void umount_tree(struct mount *mnt, enum umount_tree_flags how) 147270fbcdf4SRam Pai { 1473c003b26fSEric W. Biederman LIST_HEAD(tmp_list); 1474315fc83eSAl Viro struct mount *p; 147570fbcdf4SRam Pai 14765d88457eSEric W. Biederman if (how & UMOUNT_PROPAGATE) 14775d88457eSEric W. Biederman propagate_mount_unlock(mnt); 14785d88457eSEric W. Biederman 1479c003b26fSEric W. Biederman /* Gather the mounts to umount */ 1480590ce4bcSEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) { 1481590ce4bcSEric W. Biederman p->mnt.mnt_flags |= MNT_UMOUNT; 1482c003b26fSEric W. Biederman list_move(&p->mnt_list, &tmp_list); 1483590ce4bcSEric W. Biederman } 1484c003b26fSEric W. Biederman 1485411a938bSEric W. Biederman /* Hide the mounts from mnt_mounts */ 1486c003b26fSEric W. Biederman list_for_each_entry(p, &tmp_list, mnt_list) { 1487c003b26fSEric W. Biederman list_del_init(&p->mnt_child); 148838129a13SAl Viro } 148970fbcdf4SRam Pai 1490c003b26fSEric W. Biederman /* Add propogated mounts to the tmp_list */ 1491e819f152SEric W. Biederman if (how & UMOUNT_PROPAGATE) 14927b8a53fdSAl Viro propagate_umount(&tmp_list); 1493a05964f3SRam Pai 1494c003b26fSEric W. Biederman while (!list_empty(&tmp_list)) { 1495d2921684SEric W. Biederman struct mnt_namespace *ns; 1496ce07d891SEric W. Biederman bool disconnect; 1497c003b26fSEric W. Biederman p = list_first_entry(&tmp_list, struct mount, mnt_list); 14986776db3dSAl Viro list_del_init(&p->mnt_expire); 14991a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1500d2921684SEric W. Biederman ns = p->mnt_ns; 1501d2921684SEric W. Biederman if (ns) { 1502d2921684SEric W. Biederman ns->mounts--; 1503d2921684SEric W. Biederman __touch_mnt_namespace(ns); 1504d2921684SEric W. Biederman } 150563d37a84SAl Viro p->mnt_ns = NULL; 1506e819f152SEric W. Biederman if (how & UMOUNT_SYNC) 150748a066e7SAl Viro p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; 150887b95ce0SAl Viro 1509f2d0a123SEric W. Biederman disconnect = disconnect_mount(p, how); 1510ce07d891SEric W. Biederman 1511ce07d891SEric W. Biederman pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, 1512ce07d891SEric W. Biederman disconnect ? &unmounted : NULL); 1513676da58dSAl Viro if (mnt_has_parent(p)) { 151481b6b061SAl Viro mnt_add_count(p->mnt_parent, -1); 1515ce07d891SEric W. Biederman if (!disconnect) { 1516ce07d891SEric W. Biederman /* Don't forget about p */ 1517ce07d891SEric W. Biederman list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); 1518ce07d891SEric W. Biederman } else { 15196a46c573SEric W. Biederman umount_mnt(p); 15207c4b93d8SAl Viro } 1521ce07d891SEric W. Biederman } 15220f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 152338129a13SAl Viro } 15241da177e4SLinus Torvalds } 15251da177e4SLinus Torvalds 1526b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1527c35038beSAl Viro 15281ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 15291da177e4SLinus Torvalds { 15301ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 15311da177e4SLinus Torvalds int retval; 15321da177e4SLinus Torvalds 15331ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 15341da177e4SLinus Torvalds if (retval) 15351da177e4SLinus Torvalds return retval; 15361da177e4SLinus Torvalds 15371da177e4SLinus Torvalds /* 15381da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 15391da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 15401da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 15411da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 15421da177e4SLinus Torvalds */ 15431da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 15441ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 15451da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 15461da177e4SLinus Torvalds return -EINVAL; 15471da177e4SLinus Torvalds 1548b3e19d92SNick Piggin /* 1549b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1550b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1551b3e19d92SNick Piggin */ 1552719ea2fbSAl Viro lock_mount_hash(); 155383adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1554719ea2fbSAl Viro unlock_mount_hash(); 15551da177e4SLinus Torvalds return -EBUSY; 1556b3e19d92SNick Piggin } 1557719ea2fbSAl Viro unlock_mount_hash(); 15581da177e4SLinus Torvalds 1559863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 15601da177e4SLinus Torvalds return -EAGAIN; 15611da177e4SLinus Torvalds } 15621da177e4SLinus Torvalds 15631da177e4SLinus Torvalds /* 15641da177e4SLinus Torvalds * If we may have to abort operations to get out of this 15651da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 15661da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 15671da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 15681da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 15691da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 15701da177e4SLinus Torvalds * about for the moment. 15711da177e4SLinus Torvalds */ 15721da177e4SLinus Torvalds 157342faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 157442faad99SAl Viro sb->s_op->umount_begin(sb); 157542faad99SAl Viro } 15761da177e4SLinus Torvalds 15771da177e4SLinus Torvalds /* 15781da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 15791da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 15801da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 15811da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 15821da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 15831da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 15841da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 15851da177e4SLinus Torvalds */ 15861ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 15871da177e4SLinus Torvalds /* 15881da177e4SLinus Torvalds * Special case for "unmounting" root ... 15891da177e4SLinus Torvalds * we just try to remount it readonly. 15901da177e4SLinus Torvalds */ 1591bc6155d1SEric W. Biederman if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) 1592a1480dccSAndy Lutomirski return -EPERM; 15931da177e4SLinus Torvalds down_write(&sb->s_umount); 1594bc98a42cSDavid Howells if (!sb_rdonly(sb)) 1595e462ec50SDavid Howells retval = do_remount_sb(sb, SB_RDONLY, NULL, 0); 15961da177e4SLinus Torvalds up_write(&sb->s_umount); 15971da177e4SLinus Torvalds return retval; 15981da177e4SLinus Torvalds } 15991da177e4SLinus Torvalds 160097216be0SAl Viro namespace_lock(); 1601719ea2fbSAl Viro lock_mount_hash(); 16025addc5ddSAl Viro event++; 16031da177e4SLinus Torvalds 160448a066e7SAl Viro if (flags & MNT_DETACH) { 160548a066e7SAl Viro if (!list_empty(&mnt->mnt_list)) 1606e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE); 160748a066e7SAl Viro retval = 0; 160848a066e7SAl Viro } else { 1609b54b9be7SAl Viro shrink_submounts(mnt); 16101da177e4SLinus Torvalds retval = -EBUSY; 161148a066e7SAl Viro if (!propagate_mount_busy(mnt, 2)) { 16121a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1613e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 16141da177e4SLinus Torvalds retval = 0; 16151da177e4SLinus Torvalds } 161648a066e7SAl Viro } 1617719ea2fbSAl Viro unlock_mount_hash(); 1618e3197d83SAl Viro namespace_unlock(); 16191da177e4SLinus Torvalds return retval; 16201da177e4SLinus Torvalds } 16211da177e4SLinus Torvalds 16221da177e4SLinus Torvalds /* 162380b5dce8SEric W. Biederman * __detach_mounts - lazily unmount all mounts on the specified dentry 162480b5dce8SEric W. Biederman * 162580b5dce8SEric W. Biederman * During unlink, rmdir, and d_drop it is possible to loose the path 162680b5dce8SEric W. Biederman * to an existing mountpoint, and wind up leaking the mount. 162780b5dce8SEric W. Biederman * detach_mounts allows lazily unmounting those mounts instead of 162880b5dce8SEric W. Biederman * leaking them. 162980b5dce8SEric W. Biederman * 163080b5dce8SEric W. Biederman * The caller may hold dentry->d_inode->i_mutex. 163180b5dce8SEric W. Biederman */ 163280b5dce8SEric W. Biederman void __detach_mounts(struct dentry *dentry) 163380b5dce8SEric W. Biederman { 163480b5dce8SEric W. Biederman struct mountpoint *mp; 163580b5dce8SEric W. Biederman struct mount *mnt; 163680b5dce8SEric W. Biederman 163780b5dce8SEric W. Biederman namespace_lock(); 16383895dbf8SEric W. Biederman lock_mount_hash(); 163980b5dce8SEric W. Biederman mp = lookup_mountpoint(dentry); 1640f53e5797SEric W. Biederman if (IS_ERR_OR_NULL(mp)) 164180b5dce8SEric W. Biederman goto out_unlock; 164280b5dce8SEric W. Biederman 1643e06b933eSAndrey Ulanov event++; 164480b5dce8SEric W. Biederman while (!hlist_empty(&mp->m_list)) { 164580b5dce8SEric W. Biederman mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); 1646ce07d891SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_UMOUNT) { 1647fe78fcc8SEric W. Biederman hlist_add_head(&mnt->mnt_umount.s_list, &unmounted); 1648fe78fcc8SEric W. Biederman umount_mnt(mnt); 1649ce07d891SEric W. Biederman } 1650e0c9c0afSEric W. Biederman else umount_tree(mnt, UMOUNT_CONNECTED); 165180b5dce8SEric W. Biederman } 165280b5dce8SEric W. Biederman put_mountpoint(mp); 165380b5dce8SEric W. Biederman out_unlock: 16543895dbf8SEric W. Biederman unlock_mount_hash(); 165580b5dce8SEric W. Biederman namespace_unlock(); 165680b5dce8SEric W. Biederman } 165780b5dce8SEric W. Biederman 165880b5dce8SEric W. Biederman /* 16599b40bc90SAl Viro * Is the caller allowed to modify his namespace? 16609b40bc90SAl Viro */ 16619b40bc90SAl Viro static inline bool may_mount(void) 16629b40bc90SAl Viro { 16639b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 16649b40bc90SAl Viro } 16659b40bc90SAl Viro 16669e8925b6SJeff Layton static inline bool may_mandlock(void) 16679e8925b6SJeff Layton { 16689e8925b6SJeff Layton #ifndef CONFIG_MANDATORY_FILE_LOCKING 16699e8925b6SJeff Layton return false; 16709e8925b6SJeff Layton #endif 167195ace754SEric W. Biederman return capable(CAP_SYS_ADMIN); 16729e8925b6SJeff Layton } 16739e8925b6SJeff Layton 16749b40bc90SAl Viro /* 16751da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 16761da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 16771da177e4SLinus Torvalds * 16781da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 16791da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 16801da177e4SLinus Torvalds */ 16811da177e4SLinus Torvalds 16823a18ef5cSDominik Brodowski int ksys_umount(char __user *name, int flags) 16831da177e4SLinus Torvalds { 16842d8f3038SAl Viro struct path path; 1685900148dcSAl Viro struct mount *mnt; 16861da177e4SLinus Torvalds int retval; 1687db1f05bbSMiklos Szeredi int lookup_flags = 0; 16881da177e4SLinus Torvalds 1689db1f05bbSMiklos Szeredi if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1690db1f05bbSMiklos Szeredi return -EINVAL; 1691db1f05bbSMiklos Szeredi 16929b40bc90SAl Viro if (!may_mount()) 16939b40bc90SAl Viro return -EPERM; 16949b40bc90SAl Viro 1695db1f05bbSMiklos Szeredi if (!(flags & UMOUNT_NOFOLLOW)) 1696db1f05bbSMiklos Szeredi lookup_flags |= LOOKUP_FOLLOW; 1697db1f05bbSMiklos Szeredi 1698197df04cSAl Viro retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); 16991da177e4SLinus Torvalds if (retval) 17001da177e4SLinus Torvalds goto out; 1701900148dcSAl Viro mnt = real_mount(path.mnt); 17021da177e4SLinus Torvalds retval = -EINVAL; 17032d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 17041da177e4SLinus Torvalds goto dput_and_out; 1705143c8c91SAl Viro if (!check_mnt(mnt)) 17061da177e4SLinus Torvalds goto dput_and_out; 17075ff9d8a6SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 17085ff9d8a6SEric W. Biederman goto dput_and_out; 1709b2f5d4dcSEric W. Biederman retval = -EPERM; 1710b2f5d4dcSEric W. Biederman if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) 1711b2f5d4dcSEric W. Biederman goto dput_and_out; 17121da177e4SLinus Torvalds 1713900148dcSAl Viro retval = do_umount(mnt, flags); 17141da177e4SLinus Torvalds dput_and_out: 1715429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 17162d8f3038SAl Viro dput(path.dentry); 1717900148dcSAl Viro mntput_no_expire(mnt); 17181da177e4SLinus Torvalds out: 17191da177e4SLinus Torvalds return retval; 17201da177e4SLinus Torvalds } 17211da177e4SLinus Torvalds 17223a18ef5cSDominik Brodowski SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 17233a18ef5cSDominik Brodowski { 17243a18ef5cSDominik Brodowski return ksys_umount(name, flags); 17253a18ef5cSDominik Brodowski } 17263a18ef5cSDominik Brodowski 17271da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 17281da177e4SLinus Torvalds 17291da177e4SLinus Torvalds /* 17301da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 17311da177e4SLinus Torvalds */ 1732bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 17331da177e4SLinus Torvalds { 17343a18ef5cSDominik Brodowski return ksys_umount(name, 0); 17351da177e4SLinus Torvalds } 17361da177e4SLinus Torvalds 17371da177e4SLinus Torvalds #endif 17381da177e4SLinus Torvalds 17394ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 17408823c079SEric W. Biederman { 17414ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 1742e149ed2bSAl Viro return dentry->d_op == &ns_dentry_operations && 1743e149ed2bSAl Viro dentry->d_fsdata == &mntns_operations; 17444ce5d2b1SEric W. Biederman } 17454ce5d2b1SEric W. Biederman 174658be2825SAl Viro struct mnt_namespace *to_mnt_ns(struct ns_common *ns) 174758be2825SAl Viro { 174858be2825SAl Viro return container_of(ns, struct mnt_namespace, ns); 174958be2825SAl Viro } 175058be2825SAl Viro 17514ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 17524ce5d2b1SEric W. Biederman { 17534ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 17544ce5d2b1SEric W. Biederman * mount namespace loop? 17554ce5d2b1SEric W. Biederman */ 17564ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 17574ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 17584ce5d2b1SEric W. Biederman return false; 17594ce5d2b1SEric W. Biederman 1760f77c8014SAl Viro mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); 17618823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 17628823c079SEric W. Biederman } 17638823c079SEric W. Biederman 176487129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 176536341f64SRam Pai int flag) 17661da177e4SLinus Torvalds { 176784d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 17681da177e4SLinus Torvalds 17694ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 17704ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 17714ce5d2b1SEric W. Biederman 17724ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1773be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 17749676f0c6SRam Pai 177536341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1776be34d1a3SDavid Howells if (IS_ERR(q)) 1777be34d1a3SDavid Howells return q; 1778be34d1a3SDavid Howells 1779a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 17801da177e4SLinus Torvalds 17811da177e4SLinus Torvalds p = mnt; 17826b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1783315fc83eSAl Viro struct mount *s; 17847ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 17851da177e4SLinus Torvalds continue; 17861da177e4SLinus Torvalds 1787909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 17884ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 17894ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 17904ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 17914ce5d2b1SEric W. Biederman continue; 17924ce5d2b1SEric W. Biederman } 17934ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 17944ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 17959676f0c6SRam Pai s = skip_mnt_tree(s); 17969676f0c6SRam Pai continue; 17979676f0c6SRam Pai } 17980714a533SAl Viro while (p != s->mnt_parent) { 17990714a533SAl Viro p = p->mnt_parent; 18000714a533SAl Viro q = q->mnt_parent; 18011da177e4SLinus Torvalds } 180287129cc0SAl Viro p = s; 180384d17192SAl Viro parent = q; 180487129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1805be34d1a3SDavid Howells if (IS_ERR(q)) 1806be34d1a3SDavid Howells goto out; 1807719ea2fbSAl Viro lock_mount_hash(); 18081a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 18091064f874SEric W. Biederman attach_mnt(q, parent, p->mnt_mp); 1810719ea2fbSAl Viro unlock_mount_hash(); 18111da177e4SLinus Torvalds } 18121da177e4SLinus Torvalds } 18131da177e4SLinus Torvalds return res; 1814be34d1a3SDavid Howells out: 18151da177e4SLinus Torvalds if (res) { 1816719ea2fbSAl Viro lock_mount_hash(); 1817e819f152SEric W. Biederman umount_tree(res, UMOUNT_SYNC); 1818719ea2fbSAl Viro unlock_mount_hash(); 18191da177e4SLinus Torvalds } 1820be34d1a3SDavid Howells return q; 18211da177e4SLinus Torvalds } 18221da177e4SLinus Torvalds 1823be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1824be34d1a3SDavid Howells 1825ca71cf71SAl Viro struct vfsmount *collect_mounts(const struct path *path) 18268aec0809SAl Viro { 1827cb338d06SAl Viro struct mount *tree; 182897216be0SAl Viro namespace_lock(); 1829cd4a4017SEric W. Biederman if (!check_mnt(real_mount(path->mnt))) 1830cd4a4017SEric W. Biederman tree = ERR_PTR(-EINVAL); 1831cd4a4017SEric W. Biederman else 183287129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 183387129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1834328e6d90SAl Viro namespace_unlock(); 1835be34d1a3SDavid Howells if (IS_ERR(tree)) 183652e220d3SDan Carpenter return ERR_CAST(tree); 1837be34d1a3SDavid Howells return &tree->mnt; 18388aec0809SAl Viro } 18398aec0809SAl Viro 18408aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 18418aec0809SAl Viro { 184297216be0SAl Viro namespace_lock(); 1843719ea2fbSAl Viro lock_mount_hash(); 1844e819f152SEric W. Biederman umount_tree(real_mount(mnt), UMOUNT_SYNC); 1845719ea2fbSAl Viro unlock_mount_hash(); 18463ab6abeeSAl Viro namespace_unlock(); 18478aec0809SAl Viro } 18488aec0809SAl Viro 1849c771d683SMiklos Szeredi /** 1850c771d683SMiklos Szeredi * clone_private_mount - create a private clone of a path 1851c771d683SMiklos Szeredi * 1852c771d683SMiklos Szeredi * This creates a new vfsmount, which will be the clone of @path. The new will 1853c771d683SMiklos Szeredi * not be attached anywhere in the namespace and will be private (i.e. changes 1854c771d683SMiklos Szeredi * to the originating mount won't be propagated into this). 1855c771d683SMiklos Szeredi * 1856c771d683SMiklos Szeredi * Release with mntput(). 1857c771d683SMiklos Szeredi */ 1858ca71cf71SAl Viro struct vfsmount *clone_private_mount(const struct path *path) 1859c771d683SMiklos Szeredi { 1860c771d683SMiklos Szeredi struct mount *old_mnt = real_mount(path->mnt); 1861c771d683SMiklos Szeredi struct mount *new_mnt; 1862c771d683SMiklos Szeredi 1863c771d683SMiklos Szeredi if (IS_MNT_UNBINDABLE(old_mnt)) 1864c771d683SMiklos Szeredi return ERR_PTR(-EINVAL); 1865c771d683SMiklos Szeredi 1866c771d683SMiklos Szeredi new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); 1867c771d683SMiklos Szeredi if (IS_ERR(new_mnt)) 1868c771d683SMiklos Szeredi return ERR_CAST(new_mnt); 1869c771d683SMiklos Szeredi 1870c771d683SMiklos Szeredi return &new_mnt->mnt; 1871c771d683SMiklos Szeredi } 1872c771d683SMiklos Szeredi EXPORT_SYMBOL_GPL(clone_private_mount); 1873c771d683SMiklos Szeredi 18741f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 18751f707137SAl Viro struct vfsmount *root) 18761f707137SAl Viro { 18771a4eeaf2SAl Viro struct mount *mnt; 18781f707137SAl Viro int res = f(root, arg); 18791f707137SAl Viro if (res) 18801f707137SAl Viro return res; 18811a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 18821a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 18831f707137SAl Viro if (res) 18841f707137SAl Viro return res; 18851f707137SAl Viro } 18861f707137SAl Viro return 0; 18871f707137SAl Viro } 18881f707137SAl Viro 18894b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 1890719f5d7fSMiklos Szeredi { 1891315fc83eSAl Viro struct mount *p; 1892719f5d7fSMiklos Szeredi 1893909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1894fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 18954b8b21f4SAl Viro mnt_release_group_id(p); 1896719f5d7fSMiklos Szeredi } 1897719f5d7fSMiklos Szeredi } 1898719f5d7fSMiklos Szeredi 18994b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 1900719f5d7fSMiklos Szeredi { 1901315fc83eSAl Viro struct mount *p; 1902719f5d7fSMiklos Szeredi 1903909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1904fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 19054b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 1906719f5d7fSMiklos Szeredi if (err) { 19074b8b21f4SAl Viro cleanup_group_ids(mnt, p); 1908719f5d7fSMiklos Szeredi return err; 1909719f5d7fSMiklos Szeredi } 1910719f5d7fSMiklos Szeredi } 1911719f5d7fSMiklos Szeredi } 1912719f5d7fSMiklos Szeredi 1913719f5d7fSMiklos Szeredi return 0; 1914719f5d7fSMiklos Szeredi } 1915719f5d7fSMiklos Szeredi 1916d2921684SEric W. Biederman int count_mounts(struct mnt_namespace *ns, struct mount *mnt) 1917d2921684SEric W. Biederman { 1918d2921684SEric W. Biederman unsigned int max = READ_ONCE(sysctl_mount_max); 1919d2921684SEric W. Biederman unsigned int mounts = 0, old, pending, sum; 1920d2921684SEric W. Biederman struct mount *p; 1921d2921684SEric W. Biederman 1922d2921684SEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) 1923d2921684SEric W. Biederman mounts++; 1924d2921684SEric W. Biederman 1925d2921684SEric W. Biederman old = ns->mounts; 1926d2921684SEric W. Biederman pending = ns->pending_mounts; 1927d2921684SEric W. Biederman sum = old + pending; 1928d2921684SEric W. Biederman if ((old > sum) || 1929d2921684SEric W. Biederman (pending > sum) || 1930d2921684SEric W. Biederman (max < sum) || 1931d2921684SEric W. Biederman (mounts > (max - sum))) 1932d2921684SEric W. Biederman return -ENOSPC; 1933d2921684SEric W. Biederman 1934d2921684SEric W. Biederman ns->pending_mounts = pending + mounts; 1935d2921684SEric W. Biederman return 0; 1936d2921684SEric W. Biederman } 1937d2921684SEric W. Biederman 1938b90fa9aeSRam Pai /* 1939b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1940b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 194121444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 194221444403SRam Pai * store the parent mount and mountpoint dentry. 194321444403SRam Pai * (done when source_mnt is moved) 1944b90fa9aeSRam Pai * 1945b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1946b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 19479676f0c6SRam Pai * --------------------------------------------------------------------------- 1948b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 19499676f0c6SRam Pai * |************************************************************************** 19509676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 19519676f0c6SRam Pai * | dest | | | | | 19529676f0c6SRam Pai * | | | | | | | 19539676f0c6SRam Pai * | v | | | | | 19549676f0c6SRam Pai * |************************************************************************** 19559676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 19565afe0022SRam Pai * | | | | | | 19579676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 19589676f0c6SRam Pai * *************************************************************************** 1959b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1960b90fa9aeSRam Pai * destination mount. 1961b90fa9aeSRam Pai * 1962b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1963b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1964b90fa9aeSRam Pai * the peer group of the source mount. 1965b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1966b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1967b90fa9aeSRam Pai * mount. 19685afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 19695afe0022SRam Pai * of the destination mount and the cloned mount is made slave 19705afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 19715afe0022SRam Pai * is marked as 'shared and slave'. 19725afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 19735afe0022SRam Pai * source mount. 19745afe0022SRam Pai * 19759676f0c6SRam Pai * --------------------------------------------------------------------------- 197621444403SRam Pai * | MOVE MOUNT OPERATION | 19779676f0c6SRam Pai * |************************************************************************** 19789676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 19799676f0c6SRam Pai * | dest | | | | | 19809676f0c6SRam Pai * | | | | | | | 19819676f0c6SRam Pai * | v | | | | | 19829676f0c6SRam Pai * |************************************************************************** 19839676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 19845afe0022SRam Pai * | | | | | | 19859676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 19869676f0c6SRam Pai * *************************************************************************** 19875afe0022SRam Pai * 19885afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 19895afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 199021444403SRam Pai * (+*) the mount is moved to the destination. 19915afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 19925afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 19935afe0022SRam Pai * the mount is marked as 'shared and slave'. 19945afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1995b90fa9aeSRam Pai * 1996b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1997b90fa9aeSRam Pai * applied to each mount in the tree. 1998b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1999b90fa9aeSRam Pai * in allocations. 2000b90fa9aeSRam Pai */ 20010fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 200284d17192SAl Viro struct mount *dest_mnt, 200384d17192SAl Viro struct mountpoint *dest_mp, 200484d17192SAl Viro struct path *parent_path) 2005b90fa9aeSRam Pai { 200638129a13SAl Viro HLIST_HEAD(tree_list); 2007d2921684SEric W. Biederman struct mnt_namespace *ns = dest_mnt->mnt_ns; 20081064f874SEric W. Biederman struct mountpoint *smp; 2009315fc83eSAl Viro struct mount *child, *p; 201038129a13SAl Viro struct hlist_node *n; 2011719f5d7fSMiklos Szeredi int err; 2012b90fa9aeSRam Pai 20131064f874SEric W. Biederman /* Preallocate a mountpoint in case the new mounts need 20141064f874SEric W. Biederman * to be tucked under other mounts. 20151064f874SEric W. Biederman */ 20161064f874SEric W. Biederman smp = get_mountpoint(source_mnt->mnt.mnt_root); 20171064f874SEric W. Biederman if (IS_ERR(smp)) 20181064f874SEric W. Biederman return PTR_ERR(smp); 20191064f874SEric W. Biederman 2020d2921684SEric W. Biederman /* Is there space to add these mounts to the mount namespace? */ 2021d2921684SEric W. Biederman if (!parent_path) { 2022d2921684SEric W. Biederman err = count_mounts(ns, source_mnt); 2023d2921684SEric W. Biederman if (err) 2024d2921684SEric W. Biederman goto out; 2025d2921684SEric W. Biederman } 2026d2921684SEric W. Biederman 2027fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 20280fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 2029719f5d7fSMiklos Szeredi if (err) 2030719f5d7fSMiklos Szeredi goto out; 203184d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 2032f2ebb3a9SAl Viro lock_mount_hash(); 2033719f5d7fSMiklos Szeredi if (err) 2034719f5d7fSMiklos Szeredi goto out_cleanup_ids; 2035909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 20360f0afb1dSAl Viro set_mnt_shared(p); 20370b1b901bSAl Viro } else { 20380b1b901bSAl Viro lock_mount_hash(); 2039b90fa9aeSRam Pai } 20401a390689SAl Viro if (parent_path) { 20410fb54e50SAl Viro detach_mnt(source_mnt, parent_path); 204284d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 2043143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 204421444403SRam Pai } else { 204584d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 20461064f874SEric W. Biederman commit_tree(source_mnt); 204721444403SRam Pai } 2048b90fa9aeSRam Pai 204938129a13SAl Viro hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { 20501d6a32acSAl Viro struct mount *q; 205138129a13SAl Viro hlist_del_init(&child->mnt_hash); 20521064f874SEric W. Biederman q = __lookup_mnt(&child->mnt_parent->mnt, 20531d6a32acSAl Viro child->mnt_mountpoint); 20541064f874SEric W. Biederman if (q) 20551064f874SEric W. Biederman mnt_change_mountpoint(child, smp, q); 20561064f874SEric W. Biederman commit_tree(child); 2057b90fa9aeSRam Pai } 20581064f874SEric W. Biederman put_mountpoint(smp); 2059719ea2fbSAl Viro unlock_mount_hash(); 206099b7db7bSNick Piggin 2061b90fa9aeSRam Pai return 0; 2062719f5d7fSMiklos Szeredi 2063719f5d7fSMiklos Szeredi out_cleanup_ids: 2064f2ebb3a9SAl Viro while (!hlist_empty(&tree_list)) { 2065f2ebb3a9SAl Viro child = hlist_entry(tree_list.first, struct mount, mnt_hash); 2066d2921684SEric W. Biederman child->mnt_parent->mnt_ns->pending_mounts = 0; 2067e819f152SEric W. Biederman umount_tree(child, UMOUNT_SYNC); 2068f2ebb3a9SAl Viro } 2069f2ebb3a9SAl Viro unlock_mount_hash(); 20700fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 2071719f5d7fSMiklos Szeredi out: 2072d2921684SEric W. Biederman ns->pending_mounts = 0; 20731064f874SEric W. Biederman 20741064f874SEric W. Biederman read_seqlock_excl(&mount_lock); 20751064f874SEric W. Biederman put_mountpoint(smp); 20761064f874SEric W. Biederman read_sequnlock_excl(&mount_lock); 20771064f874SEric W. Biederman 2078719f5d7fSMiklos Szeredi return err; 2079b90fa9aeSRam Pai } 2080b90fa9aeSRam Pai 208184d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 2082b12cea91SAl Viro { 2083b12cea91SAl Viro struct vfsmount *mnt; 208484d17192SAl Viro struct dentry *dentry = path->dentry; 2085b12cea91SAl Viro retry: 20865955102cSAl Viro inode_lock(dentry->d_inode); 208784d17192SAl Viro if (unlikely(cant_mount(dentry))) { 20885955102cSAl Viro inode_unlock(dentry->d_inode); 208984d17192SAl Viro return ERR_PTR(-ENOENT); 2090b12cea91SAl Viro } 209197216be0SAl Viro namespace_lock(); 2092b12cea91SAl Viro mnt = lookup_mnt(path); 209384d17192SAl Viro if (likely(!mnt)) { 20943895dbf8SEric W. Biederman struct mountpoint *mp = get_mountpoint(dentry); 209584d17192SAl Viro if (IS_ERR(mp)) { 209697216be0SAl Viro namespace_unlock(); 20975955102cSAl Viro inode_unlock(dentry->d_inode); 209884d17192SAl Viro return mp; 209984d17192SAl Viro } 210084d17192SAl Viro return mp; 210184d17192SAl Viro } 210297216be0SAl Viro namespace_unlock(); 21035955102cSAl Viro inode_unlock(path->dentry->d_inode); 2104b12cea91SAl Viro path_put(path); 2105b12cea91SAl Viro path->mnt = mnt; 210684d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 2107b12cea91SAl Viro goto retry; 2108b12cea91SAl Viro } 2109b12cea91SAl Viro 211084d17192SAl Viro static void unlock_mount(struct mountpoint *where) 2111b12cea91SAl Viro { 211284d17192SAl Viro struct dentry *dentry = where->m_dentry; 21133895dbf8SEric W. Biederman 21143895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 211584d17192SAl Viro put_mountpoint(where); 21163895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 21173895dbf8SEric W. Biederman 2118328e6d90SAl Viro namespace_unlock(); 21195955102cSAl Viro inode_unlock(dentry->d_inode); 2120b12cea91SAl Viro } 2121b12cea91SAl Viro 212284d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 21231da177e4SLinus Torvalds { 2124e462ec50SDavid Howells if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) 21251da177e4SLinus Torvalds return -EINVAL; 21261da177e4SLinus Torvalds 2127e36cb0b8SDavid Howells if (d_is_dir(mp->m_dentry) != 2128e36cb0b8SDavid Howells d_is_dir(mnt->mnt.mnt_root)) 21291da177e4SLinus Torvalds return -ENOTDIR; 21301da177e4SLinus Torvalds 213184d17192SAl Viro return attach_recursive_mnt(mnt, p, mp, NULL); 21321da177e4SLinus Torvalds } 21331da177e4SLinus Torvalds 21341da177e4SLinus Torvalds /* 21357a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 21367a2e8a8fSValerie Aurora */ 21377a2e8a8fSValerie Aurora 2138e462ec50SDavid Howells static int flags_to_propagation_type(int ms_flags) 21397a2e8a8fSValerie Aurora { 2140e462ec50SDavid Howells int type = ms_flags & ~(MS_REC | MS_SILENT); 21417a2e8a8fSValerie Aurora 21427a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 21437a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 21447a2e8a8fSValerie Aurora return 0; 21457a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 21467a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 21477a2e8a8fSValerie Aurora return 0; 21487a2e8a8fSValerie Aurora return type; 21497a2e8a8fSValerie Aurora } 21507a2e8a8fSValerie Aurora 21517a2e8a8fSValerie Aurora /* 215207b20889SRam Pai * recursively change the type of the mountpoint. 215307b20889SRam Pai */ 2154e462ec50SDavid Howells static int do_change_type(struct path *path, int ms_flags) 215507b20889SRam Pai { 2156315fc83eSAl Viro struct mount *m; 21574b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 2158e462ec50SDavid Howells int recurse = ms_flags & MS_REC; 21597a2e8a8fSValerie Aurora int type; 2160719f5d7fSMiklos Szeredi int err = 0; 216107b20889SRam Pai 21622d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 216307b20889SRam Pai return -EINVAL; 216407b20889SRam Pai 2165e462ec50SDavid Howells type = flags_to_propagation_type(ms_flags); 21667a2e8a8fSValerie Aurora if (!type) 21677a2e8a8fSValerie Aurora return -EINVAL; 21687a2e8a8fSValerie Aurora 216997216be0SAl Viro namespace_lock(); 2170719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 2171719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 2172719f5d7fSMiklos Szeredi if (err) 2173719f5d7fSMiklos Szeredi goto out_unlock; 2174719f5d7fSMiklos Szeredi } 2175719f5d7fSMiklos Szeredi 2176719ea2fbSAl Viro lock_mount_hash(); 2177909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 21780f0afb1dSAl Viro change_mnt_propagation(m, type); 2179719ea2fbSAl Viro unlock_mount_hash(); 2180719f5d7fSMiklos Szeredi 2181719f5d7fSMiklos Szeredi out_unlock: 218297216be0SAl Viro namespace_unlock(); 2183719f5d7fSMiklos Szeredi return err; 218407b20889SRam Pai } 218507b20889SRam Pai 21865ff9d8a6SEric W. Biederman static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 21875ff9d8a6SEric W. Biederman { 21885ff9d8a6SEric W. Biederman struct mount *child; 21895ff9d8a6SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 21905ff9d8a6SEric W. Biederman if (!is_subdir(child->mnt_mountpoint, dentry)) 21915ff9d8a6SEric W. Biederman continue; 21925ff9d8a6SEric W. Biederman 21935ff9d8a6SEric W. Biederman if (child->mnt.mnt_flags & MNT_LOCKED) 21945ff9d8a6SEric W. Biederman return true; 21955ff9d8a6SEric W. Biederman } 21965ff9d8a6SEric W. Biederman return false; 21975ff9d8a6SEric W. Biederman } 21985ff9d8a6SEric W. Biederman 219907b20889SRam Pai /* 22001da177e4SLinus Torvalds * do loopback mount. 22011da177e4SLinus Torvalds */ 2202808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 22032dafe1c4SEric Sandeen int recurse) 22041da177e4SLinus Torvalds { 22052d92ab3cSAl Viro struct path old_path; 220684d17192SAl Viro struct mount *mnt = NULL, *old, *parent; 220784d17192SAl Viro struct mountpoint *mp; 220857eccb83SAl Viro int err; 22091da177e4SLinus Torvalds if (!old_name || !*old_name) 22101da177e4SLinus Torvalds return -EINVAL; 2211815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 22121da177e4SLinus Torvalds if (err) 22131da177e4SLinus Torvalds return err; 22141da177e4SLinus Torvalds 22158823c079SEric W. Biederman err = -EINVAL; 22164ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 22178823c079SEric W. Biederman goto out; 22188823c079SEric W. Biederman 221984d17192SAl Viro mp = lock_mount(path); 222084d17192SAl Viro err = PTR_ERR(mp); 222184d17192SAl Viro if (IS_ERR(mp)) 22229676f0c6SRam Pai goto out; 22239676f0c6SRam Pai 222487129cc0SAl Viro old = real_mount(old_path.mnt); 222584d17192SAl Viro parent = real_mount(path->mnt); 222687129cc0SAl Viro 2227b12cea91SAl Viro err = -EINVAL; 2228fc7be130SAl Viro if (IS_MNT_UNBINDABLE(old)) 2229b12cea91SAl Viro goto out2; 2230b12cea91SAl Viro 2231e149ed2bSAl Viro if (!check_mnt(parent)) 2232e149ed2bSAl Viro goto out2; 2233e149ed2bSAl Viro 2234e149ed2bSAl Viro if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations) 2235b12cea91SAl Viro goto out2; 2236ccd48bc7SAl Viro 22375ff9d8a6SEric W. Biederman if (!recurse && has_locked_children(old, old_path.dentry)) 22385ff9d8a6SEric W. Biederman goto out2; 22395ff9d8a6SEric W. Biederman 22401da177e4SLinus Torvalds if (recurse) 22414ce5d2b1SEric W. Biederman mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); 22421da177e4SLinus Torvalds else 224387129cc0SAl Viro mnt = clone_mnt(old, old_path.dentry, 0); 22441da177e4SLinus Torvalds 2245be34d1a3SDavid Howells if (IS_ERR(mnt)) { 2246be34d1a3SDavid Howells err = PTR_ERR(mnt); 2247e9c5d8a5SAndrey Vagin goto out2; 2248be34d1a3SDavid Howells } 2249ccd48bc7SAl Viro 22505ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags &= ~MNT_LOCKED; 22515ff9d8a6SEric W. Biederman 225284d17192SAl Viro err = graft_tree(mnt, parent, mp); 22531da177e4SLinus Torvalds if (err) { 2254719ea2fbSAl Viro lock_mount_hash(); 2255e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_SYNC); 2256719ea2fbSAl Viro unlock_mount_hash(); 22575b83d2c5SRam Pai } 2258b12cea91SAl Viro out2: 225984d17192SAl Viro unlock_mount(mp); 2260ccd48bc7SAl Viro out: 22612d92ab3cSAl Viro path_put(&old_path); 22621da177e4SLinus Torvalds return err; 22631da177e4SLinus Torvalds } 22641da177e4SLinus Torvalds 22652e4b7fcdSDave Hansen static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 22662e4b7fcdSDave Hansen { 22672e4b7fcdSDave Hansen int error = 0; 22682e4b7fcdSDave Hansen int readonly_request = 0; 22692e4b7fcdSDave Hansen 22702e4b7fcdSDave Hansen if (ms_flags & MS_RDONLY) 22712e4b7fcdSDave Hansen readonly_request = 1; 22722e4b7fcdSDave Hansen if (readonly_request == __mnt_is_readonly(mnt)) 22732e4b7fcdSDave Hansen return 0; 22742e4b7fcdSDave Hansen 22752e4b7fcdSDave Hansen if (readonly_request) 227683adc753SAl Viro error = mnt_make_readonly(real_mount(mnt)); 22772e4b7fcdSDave Hansen else 227883adc753SAl Viro __mnt_unmake_readonly(real_mount(mnt)); 22792e4b7fcdSDave Hansen return error; 22802e4b7fcdSDave Hansen } 22812e4b7fcdSDave Hansen 22821da177e4SLinus Torvalds /* 22831da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 22841da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 22851da177e4SLinus Torvalds * on it - tough luck. 22861da177e4SLinus Torvalds */ 2287e462ec50SDavid Howells static int do_remount(struct path *path, int ms_flags, int sb_flags, 2288e462ec50SDavid Howells int mnt_flags, void *data) 22891da177e4SLinus Torvalds { 22901da177e4SLinus Torvalds int err; 22912d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 2292143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 22931da177e4SLinus Torvalds 2294143c8c91SAl Viro if (!check_mnt(mnt)) 22951da177e4SLinus Torvalds return -EINVAL; 22961da177e4SLinus Torvalds 22972d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 22981da177e4SLinus Torvalds return -EINVAL; 22991da177e4SLinus Torvalds 230007b64558SEric W. Biederman /* Don't allow changing of locked mnt flags. 230107b64558SEric W. Biederman * 230207b64558SEric W. Biederman * No locks need to be held here while testing the various 230307b64558SEric W. Biederman * MNT_LOCK flags because those flags can never be cleared 230407b64558SEric W. Biederman * once they are set. 230507b64558SEric W. Biederman */ 230607b64558SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) && 230707b64558SEric W. Biederman !(mnt_flags & MNT_READONLY)) { 230807b64558SEric W. Biederman return -EPERM; 230907b64558SEric W. Biederman } 23109566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) && 23119566d674SEric W. Biederman !(mnt_flags & MNT_NODEV)) { 23129566d674SEric W. Biederman return -EPERM; 23139566d674SEric W. Biederman } 23149566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) && 23159566d674SEric W. Biederman !(mnt_flags & MNT_NOSUID)) { 23169566d674SEric W. Biederman return -EPERM; 23179566d674SEric W. Biederman } 23189566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) && 23199566d674SEric W. Biederman !(mnt_flags & MNT_NOEXEC)) { 23209566d674SEric W. Biederman return -EPERM; 23219566d674SEric W. Biederman } 23229566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) && 23239566d674SEric W. Biederman ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) { 23249566d674SEric W. Biederman return -EPERM; 23259566d674SEric W. Biederman } 23269566d674SEric W. Biederman 2327ff36fe2cSEric Paris err = security_sb_remount(sb, data); 2328ff36fe2cSEric Paris if (err) 2329ff36fe2cSEric Paris return err; 2330ff36fe2cSEric Paris 23311da177e4SLinus Torvalds down_write(&sb->s_umount); 2332e462ec50SDavid Howells if (ms_flags & MS_BIND) 2333e462ec50SDavid Howells err = change_mount_flags(path->mnt, ms_flags); 2334bc6155d1SEric W. Biederman else if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) 233557eccb83SAl Viro err = -EPERM; 23364aa98cf7SAl Viro else 2337e462ec50SDavid Howells err = do_remount_sb(sb, sb_flags, data, 0); 23387b43a79fSAl Viro if (!err) { 2339719ea2fbSAl Viro lock_mount_hash(); 2340a6138db8SEric W. Biederman mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; 2341143c8c91SAl Viro mnt->mnt.mnt_flags = mnt_flags; 2342143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2343719ea2fbSAl Viro unlock_mount_hash(); 23440e55a7ccSDan Williams } 23456339dab8SAl Viro up_write(&sb->s_umount); 23461da177e4SLinus Torvalds return err; 23471da177e4SLinus Torvalds } 23481da177e4SLinus Torvalds 2349cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 23509676f0c6SRam Pai { 2351315fc83eSAl Viro struct mount *p; 2352909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2353fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 23549676f0c6SRam Pai return 1; 23559676f0c6SRam Pai } 23569676f0c6SRam Pai return 0; 23579676f0c6SRam Pai } 23589676f0c6SRam Pai 2359808d4e3cSAl Viro static int do_move_mount(struct path *path, const char *old_name) 23601da177e4SLinus Torvalds { 23612d92ab3cSAl Viro struct path old_path, parent_path; 2362676da58dSAl Viro struct mount *p; 23630fb54e50SAl Viro struct mount *old; 236484d17192SAl Viro struct mountpoint *mp; 236557eccb83SAl Viro int err; 23661da177e4SLinus Torvalds if (!old_name || !*old_name) 23671da177e4SLinus Torvalds return -EINVAL; 23682d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 23691da177e4SLinus Torvalds if (err) 23701da177e4SLinus Torvalds return err; 23711da177e4SLinus Torvalds 237284d17192SAl Viro mp = lock_mount(path); 237384d17192SAl Viro err = PTR_ERR(mp); 237484d17192SAl Viro if (IS_ERR(mp)) 2375cc53ce53SDavid Howells goto out; 2376cc53ce53SDavid Howells 2377143c8c91SAl Viro old = real_mount(old_path.mnt); 2378fc7be130SAl Viro p = real_mount(path->mnt); 2379143c8c91SAl Viro 23801da177e4SLinus Torvalds err = -EINVAL; 2381fc7be130SAl Viro if (!check_mnt(p) || !check_mnt(old)) 23821da177e4SLinus Torvalds goto out1; 23831da177e4SLinus Torvalds 23845ff9d8a6SEric W. Biederman if (old->mnt.mnt_flags & MNT_LOCKED) 23855ff9d8a6SEric W. Biederman goto out1; 23865ff9d8a6SEric W. Biederman 23871da177e4SLinus Torvalds err = -EINVAL; 23882d92ab3cSAl Viro if (old_path.dentry != old_path.mnt->mnt_root) 238921444403SRam Pai goto out1; 23901da177e4SLinus Torvalds 2391676da58dSAl Viro if (!mnt_has_parent(old)) 239221444403SRam Pai goto out1; 23931da177e4SLinus Torvalds 2394e36cb0b8SDavid Howells if (d_is_dir(path->dentry) != 2395e36cb0b8SDavid Howells d_is_dir(old_path.dentry)) 239621444403SRam Pai goto out1; 239721444403SRam Pai /* 239821444403SRam Pai * Don't move a mount residing in a shared parent. 239921444403SRam Pai */ 2400fc7be130SAl Viro if (IS_MNT_SHARED(old->mnt_parent)) 240121444403SRam Pai goto out1; 24029676f0c6SRam Pai /* 24039676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 24049676f0c6SRam Pai * mount which is shared. 24059676f0c6SRam Pai */ 2406fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 24079676f0c6SRam Pai goto out1; 24081da177e4SLinus Torvalds err = -ELOOP; 2409fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 2410676da58dSAl Viro if (p == old) 241121444403SRam Pai goto out1; 24121da177e4SLinus Torvalds 241384d17192SAl Viro err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); 24144ac91378SJan Blunck if (err) 241521444403SRam Pai goto out1; 24161da177e4SLinus Torvalds 24171da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 24181da177e4SLinus Torvalds * automatically */ 24196776db3dSAl Viro list_del_init(&old->mnt_expire); 24201da177e4SLinus Torvalds out1: 242184d17192SAl Viro unlock_mount(mp); 24221da177e4SLinus Torvalds out: 24231da177e4SLinus Torvalds if (!err) 24241a390689SAl Viro path_put(&parent_path); 24252d92ab3cSAl Viro path_put(&old_path); 24261da177e4SLinus Torvalds return err; 24271da177e4SLinus Torvalds } 24281da177e4SLinus Torvalds 24299d412a43SAl Viro static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) 24309d412a43SAl Viro { 24319d412a43SAl Viro int err; 24329d412a43SAl Viro const char *subtype = strchr(fstype, '.'); 24339d412a43SAl Viro if (subtype) { 24349d412a43SAl Viro subtype++; 24359d412a43SAl Viro err = -EINVAL; 24369d412a43SAl Viro if (!subtype[0]) 24379d412a43SAl Viro goto err; 24389d412a43SAl Viro } else 24399d412a43SAl Viro subtype = ""; 24409d412a43SAl Viro 24419d412a43SAl Viro mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); 24429d412a43SAl Viro err = -ENOMEM; 24439d412a43SAl Viro if (!mnt->mnt_sb->s_subtype) 24449d412a43SAl Viro goto err; 24459d412a43SAl Viro return mnt; 24469d412a43SAl Viro 24479d412a43SAl Viro err: 24489d412a43SAl Viro mntput(mnt); 24499d412a43SAl Viro return ERR_PTR(err); 24509d412a43SAl Viro } 24519d412a43SAl Viro 24529d412a43SAl Viro /* 24539d412a43SAl Viro * add a mount into a namespace's mount tree 24549d412a43SAl Viro */ 245595bc5f25SAl Viro static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) 24569d412a43SAl Viro { 245784d17192SAl Viro struct mountpoint *mp; 245884d17192SAl Viro struct mount *parent; 24599d412a43SAl Viro int err; 24609d412a43SAl Viro 2461f2ebb3a9SAl Viro mnt_flags &= ~MNT_INTERNAL_FLAGS; 24629d412a43SAl Viro 246384d17192SAl Viro mp = lock_mount(path); 246484d17192SAl Viro if (IS_ERR(mp)) 246584d17192SAl Viro return PTR_ERR(mp); 24669d412a43SAl Viro 246784d17192SAl Viro parent = real_mount(path->mnt); 24689d412a43SAl Viro err = -EINVAL; 246984d17192SAl Viro if (unlikely(!check_mnt(parent))) { 2470156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 2471156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 24729d412a43SAl Viro goto unlock; 2473156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 247484d17192SAl Viro if (!parent->mnt_ns) 2475156cacb1SAl Viro goto unlock; 2476156cacb1SAl Viro } 24779d412a43SAl Viro 24789d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 24799d412a43SAl Viro err = -EBUSY; 248095bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 24819d412a43SAl Viro path->mnt->mnt_root == path->dentry) 24829d412a43SAl Viro goto unlock; 24839d412a43SAl Viro 24849d412a43SAl Viro err = -EINVAL; 2485e36cb0b8SDavid Howells if (d_is_symlink(newmnt->mnt.mnt_root)) 24869d412a43SAl Viro goto unlock; 24879d412a43SAl Viro 248895bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 248984d17192SAl Viro err = graft_tree(newmnt, parent, mp); 24909d412a43SAl Viro 24919d412a43SAl Viro unlock: 249284d17192SAl Viro unlock_mount(mp); 24939d412a43SAl Viro return err; 24949d412a43SAl Viro } 2495b1e75df4SAl Viro 24968654df4eSEric W. Biederman static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags); 24971b852bceSEric W. Biederman 24981da177e4SLinus Torvalds /* 24991da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 25001da177e4SLinus Torvalds * namespace's tree 25011da177e4SLinus Torvalds */ 2502e462ec50SDavid Howells static int do_new_mount(struct path *path, const char *fstype, int sb_flags, 2503808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 25041da177e4SLinus Torvalds { 25050c55cfc4SEric W. Biederman struct file_system_type *type; 25061da177e4SLinus Torvalds struct vfsmount *mnt; 250715f9a3f3SAl Viro int err; 25081da177e4SLinus Torvalds 25090c55cfc4SEric W. Biederman if (!fstype) 25101da177e4SLinus Torvalds return -EINVAL; 25111da177e4SLinus Torvalds 25120c55cfc4SEric W. Biederman type = get_fs_type(fstype); 25130c55cfc4SEric W. Biederman if (!type) 25140c55cfc4SEric W. Biederman return -ENODEV; 25150c55cfc4SEric W. Biederman 2516e462ec50SDavid Howells mnt = vfs_kern_mount(type, sb_flags, name, data); 25170c55cfc4SEric W. Biederman if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && 25180c55cfc4SEric W. Biederman !mnt->mnt_sb->s_subtype) 25190c55cfc4SEric W. Biederman mnt = fs_set_subtype(mnt, fstype); 25200c55cfc4SEric W. Biederman 25210c55cfc4SEric W. Biederman put_filesystem(type); 25221da177e4SLinus Torvalds if (IS_ERR(mnt)) 25231da177e4SLinus Torvalds return PTR_ERR(mnt); 25241da177e4SLinus Torvalds 25258654df4eSEric W. Biederman if (mount_too_revealing(mnt, &mnt_flags)) { 25268654df4eSEric W. Biederman mntput(mnt); 25278654df4eSEric W. Biederman return -EPERM; 25288654df4eSEric W. Biederman } 25298654df4eSEric W. Biederman 253095bc5f25SAl Viro err = do_add_mount(real_mount(mnt), path, mnt_flags); 253115f9a3f3SAl Viro if (err) 253215f9a3f3SAl Viro mntput(mnt); 253315f9a3f3SAl Viro return err; 25341da177e4SLinus Torvalds } 25351da177e4SLinus Torvalds 253619a167afSAl Viro int finish_automount(struct vfsmount *m, struct path *path) 253719a167afSAl Viro { 25386776db3dSAl Viro struct mount *mnt = real_mount(m); 253919a167afSAl Viro int err; 254019a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 254119a167afSAl Viro * expired before we get a chance to add it 254219a167afSAl Viro */ 25436776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 254419a167afSAl Viro 254519a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 254619a167afSAl Viro m->mnt_root == path->dentry) { 2547b1e75df4SAl Viro err = -ELOOP; 2548b1e75df4SAl Viro goto fail; 254919a167afSAl Viro } 255019a167afSAl Viro 255195bc5f25SAl Viro err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 2552b1e75df4SAl Viro if (!err) 2553b1e75df4SAl Viro return 0; 2554b1e75df4SAl Viro fail: 2555b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 25566776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 255797216be0SAl Viro namespace_lock(); 25586776db3dSAl Viro list_del_init(&mnt->mnt_expire); 255997216be0SAl Viro namespace_unlock(); 256019a167afSAl Viro } 2561b1e75df4SAl Viro mntput(m); 2562b1e75df4SAl Viro mntput(m); 256319a167afSAl Viro return err; 256419a167afSAl Viro } 256519a167afSAl Viro 2566ea5b778aSDavid Howells /** 2567ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 2568ea5b778aSDavid Howells * @mnt: The mount to list. 2569ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 2570ea5b778aSDavid Howells */ 2571ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 2572ea5b778aSDavid Howells { 257397216be0SAl Viro namespace_lock(); 2574ea5b778aSDavid Howells 25756776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 2576ea5b778aSDavid Howells 257797216be0SAl Viro namespace_unlock(); 2578ea5b778aSDavid Howells } 2579ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 2580ea5b778aSDavid Howells 2581ea5b778aSDavid Howells /* 25821da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 25831da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 25841da177e4SLinus Torvalds * here 25851da177e4SLinus Torvalds */ 25861da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 25871da177e4SLinus Torvalds { 2588761d5c38SAl Viro struct mount *mnt, *next; 25891da177e4SLinus Torvalds LIST_HEAD(graveyard); 25901da177e4SLinus Torvalds 25911da177e4SLinus Torvalds if (list_empty(mounts)) 25921da177e4SLinus Torvalds return; 25931da177e4SLinus Torvalds 259497216be0SAl Viro namespace_lock(); 2595719ea2fbSAl Viro lock_mount_hash(); 25961da177e4SLinus Torvalds 25971da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 25981da177e4SLinus Torvalds * following criteria: 25991da177e4SLinus Torvalds * - only referenced by its parent vfsmount 26001da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 26011da177e4SLinus Torvalds * cleared by mntput()) 26021da177e4SLinus Torvalds */ 26036776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 2604863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 26051ab59738SAl Viro propagate_mount_busy(mnt, 1)) 26061da177e4SLinus Torvalds continue; 26076776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 26081da177e4SLinus Torvalds } 2609bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 26106776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 2611143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2612e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2613bcc5c7d2SAl Viro } 2614719ea2fbSAl Viro unlock_mount_hash(); 26153ab6abeeSAl Viro namespace_unlock(); 26161da177e4SLinus Torvalds } 26171da177e4SLinus Torvalds 26181da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 26191da177e4SLinus Torvalds 26201da177e4SLinus Torvalds /* 26215528f911STrond Myklebust * Ripoff of 'select_parent()' 26225528f911STrond Myklebust * 26235528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 26245528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 26255528f911STrond Myklebust */ 2626692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 26275528f911STrond Myklebust { 2628692afc31SAl Viro struct mount *this_parent = parent; 26295528f911STrond Myklebust struct list_head *next; 26305528f911STrond Myklebust int found = 0; 26315528f911STrond Myklebust 26325528f911STrond Myklebust repeat: 26336b41d536SAl Viro next = this_parent->mnt_mounts.next; 26345528f911STrond Myklebust resume: 26356b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 26365528f911STrond Myklebust struct list_head *tmp = next; 26376b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 26385528f911STrond Myklebust 26395528f911STrond Myklebust next = tmp->next; 2640692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 26415528f911STrond Myklebust continue; 26425528f911STrond Myklebust /* 26435528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 26445528f911STrond Myklebust */ 26456b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 26465528f911STrond Myklebust this_parent = mnt; 26475528f911STrond Myklebust goto repeat; 26485528f911STrond Myklebust } 26495528f911STrond Myklebust 26501ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 26516776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 26525528f911STrond Myklebust found++; 26535528f911STrond Myklebust } 26545528f911STrond Myklebust } 26555528f911STrond Myklebust /* 26565528f911STrond Myklebust * All done at this level ... ascend and resume the search 26575528f911STrond Myklebust */ 26585528f911STrond Myklebust if (this_parent != parent) { 26596b41d536SAl Viro next = this_parent->mnt_child.next; 26600714a533SAl Viro this_parent = this_parent->mnt_parent; 26615528f911STrond Myklebust goto resume; 26625528f911STrond Myklebust } 26635528f911STrond Myklebust return found; 26645528f911STrond Myklebust } 26655528f911STrond Myklebust 26665528f911STrond Myklebust /* 26675528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 26685528f911STrond Myklebust * submounts of a specific parent mountpoint 266999b7db7bSNick Piggin * 267048a066e7SAl Viro * mount_lock must be held for write 26715528f911STrond Myklebust */ 2672b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 26735528f911STrond Myklebust { 26745528f911STrond Myklebust LIST_HEAD(graveyard); 2675761d5c38SAl Viro struct mount *m; 26765528f911STrond Myklebust 26775528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 2678c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 2679bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 2680761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 26816776db3dSAl Viro mnt_expire); 2682143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 2683e819f152SEric W. Biederman umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2684bcc5c7d2SAl Viro } 2685bcc5c7d2SAl Viro } 26865528f911STrond Myklebust } 26875528f911STrond Myklebust 26885528f911STrond Myklebust /* 26891da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 26901da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 26911da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 26921da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 26931da177e4SLinus Torvalds */ 2694b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 2695b58fed8bSRam Pai unsigned long n) 26961da177e4SLinus Torvalds { 26971da177e4SLinus Torvalds char *t = to; 26981da177e4SLinus Torvalds const char __user *f = from; 26991da177e4SLinus Torvalds char c; 27001da177e4SLinus Torvalds 27011da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 27021da177e4SLinus Torvalds return n; 27031da177e4SLinus Torvalds 27041da177e4SLinus Torvalds while (n) { 27051da177e4SLinus Torvalds if (__get_user(c, f)) { 27061da177e4SLinus Torvalds memset(t, 0, n); 27071da177e4SLinus Torvalds break; 27081da177e4SLinus Torvalds } 27091da177e4SLinus Torvalds *t++ = c; 27101da177e4SLinus Torvalds f++; 27111da177e4SLinus Torvalds n--; 27121da177e4SLinus Torvalds } 27131da177e4SLinus Torvalds return n; 27141da177e4SLinus Torvalds } 27151da177e4SLinus Torvalds 2716b40ef869SAl Viro void *copy_mount_options(const void __user * data) 27171da177e4SLinus Torvalds { 27181da177e4SLinus Torvalds int i; 27191da177e4SLinus Torvalds unsigned long size; 2720b40ef869SAl Viro char *copy; 27211da177e4SLinus Torvalds 27221da177e4SLinus Torvalds if (!data) 2723b40ef869SAl Viro return NULL; 27241da177e4SLinus Torvalds 2725b40ef869SAl Viro copy = kmalloc(PAGE_SIZE, GFP_KERNEL); 2726b40ef869SAl Viro if (!copy) 2727b40ef869SAl Viro return ERR_PTR(-ENOMEM); 27281da177e4SLinus Torvalds 27291da177e4SLinus Torvalds /* We only care that *some* data at the address the user 27301da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 27311da177e4SLinus Torvalds * the remainder of the page. 27321da177e4SLinus Torvalds */ 27331da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 27341da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 27351da177e4SLinus Torvalds if (size > PAGE_SIZE) 27361da177e4SLinus Torvalds size = PAGE_SIZE; 27371da177e4SLinus Torvalds 2738b40ef869SAl Viro i = size - exact_copy_from_user(copy, data, size); 27391da177e4SLinus Torvalds if (!i) { 2740b40ef869SAl Viro kfree(copy); 2741b40ef869SAl Viro return ERR_PTR(-EFAULT); 27421da177e4SLinus Torvalds } 27431da177e4SLinus Torvalds if (i != PAGE_SIZE) 2744b40ef869SAl Viro memset(copy + i, 0, PAGE_SIZE - i); 2745b40ef869SAl Viro return copy; 27461da177e4SLinus Torvalds } 27471da177e4SLinus Torvalds 2748b8850d1fSTim Gardner char *copy_mount_string(const void __user *data) 2749eca6f534SVegard Nossum { 2750b8850d1fSTim Gardner return data ? strndup_user(data, PAGE_SIZE) : NULL; 2751eca6f534SVegard Nossum } 2752eca6f534SVegard Nossum 27531da177e4SLinus Torvalds /* 27541da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 27551da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 27561da177e4SLinus Torvalds * 27571da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 27581da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 27591da177e4SLinus Torvalds * information (or be NULL). 27601da177e4SLinus Torvalds * 27611da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 27621da177e4SLinus Torvalds * When the flags word was introduced its top half was required 27631da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 27641da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 27651da177e4SLinus Torvalds * and must be discarded. 27661da177e4SLinus Torvalds */ 27675e6123f3SSeunghun Lee long do_mount(const char *dev_name, const char __user *dir_name, 2768808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 27691da177e4SLinus Torvalds { 27702d92ab3cSAl Viro struct path path; 2771e462ec50SDavid Howells unsigned int mnt_flags = 0, sb_flags; 27721da177e4SLinus Torvalds int retval = 0; 27731da177e4SLinus Torvalds 27741da177e4SLinus Torvalds /* Discard magic */ 27751da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 27761da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 27771da177e4SLinus Torvalds 27781da177e4SLinus Torvalds /* Basic sanity checks */ 27791da177e4SLinus Torvalds if (data_page) 27801da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 27811da177e4SLinus Torvalds 2782e462ec50SDavid Howells if (flags & MS_NOUSER) 2783e462ec50SDavid Howells return -EINVAL; 2784e462ec50SDavid Howells 2785a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 27865e6123f3SSeunghun Lee retval = user_path(dir_name, &path); 2787a27ab9f2STetsuo Handa if (retval) 2788a27ab9f2STetsuo Handa return retval; 2789a27ab9f2STetsuo Handa 2790a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 2791a27ab9f2STetsuo Handa type_page, flags, data_page); 27920d5cadb8SAl Viro if (!retval && !may_mount()) 27930d5cadb8SAl Viro retval = -EPERM; 2794e462ec50SDavid Howells if (!retval && (flags & SB_MANDLOCK) && !may_mandlock()) 27959e8925b6SJeff Layton retval = -EPERM; 2796a27ab9f2STetsuo Handa if (retval) 2797a27ab9f2STetsuo Handa goto dput_out; 2798a27ab9f2STetsuo Handa 2799613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 2800613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 28010a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 28020a1c01c9SMatthew Garrett 28031da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 28041da177e4SLinus Torvalds if (flags & MS_NOSUID) 28051da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 28061da177e4SLinus Torvalds if (flags & MS_NODEV) 28071da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 28081da177e4SLinus Torvalds if (flags & MS_NOEXEC) 28091da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 2810fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 2811fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 2812fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 2813fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 2814d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 2815d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 2816a9e5b732SDavid Howells if (flags & MS_RDONLY) 28172e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 2818fc33a7bbSChristoph Hellwig 2819ffbc6f0eSEric W. Biederman /* The default atime for remount is preservation */ 2820ffbc6f0eSEric W. Biederman if ((flags & MS_REMOUNT) && 2821ffbc6f0eSEric W. Biederman ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | 2822ffbc6f0eSEric W. Biederman MS_STRICTATIME)) == 0)) { 2823ffbc6f0eSEric W. Biederman mnt_flags &= ~MNT_ATIME_MASK; 2824ffbc6f0eSEric W. Biederman mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; 2825ffbc6f0eSEric W. Biederman } 2826ffbc6f0eSEric W. Biederman 2827e462ec50SDavid Howells sb_flags = flags & (SB_RDONLY | 2828e462ec50SDavid Howells SB_SYNCHRONOUS | 2829e462ec50SDavid Howells SB_MANDLOCK | 2830e462ec50SDavid Howells SB_DIRSYNC | 2831e462ec50SDavid Howells SB_SILENT | 2832917086ffSMimi Zohar SB_POSIXACL | 2833d7ee9469SMarkus Trippelsdorf SB_LAZYTIME | 2834917086ffSMimi Zohar SB_I_VERSION); 28351da177e4SLinus Torvalds 28361da177e4SLinus Torvalds if (flags & MS_REMOUNT) 2837e462ec50SDavid Howells retval = do_remount(&path, flags, sb_flags, mnt_flags, 28381da177e4SLinus Torvalds data_page); 28391da177e4SLinus Torvalds else if (flags & MS_BIND) 28402d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 28419676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 28422d92ab3cSAl Viro retval = do_change_type(&path, flags); 28431da177e4SLinus Torvalds else if (flags & MS_MOVE) 28442d92ab3cSAl Viro retval = do_move_mount(&path, dev_name); 28451da177e4SLinus Torvalds else 2846e462ec50SDavid Howells retval = do_new_mount(&path, type_page, sb_flags, mnt_flags, 28471da177e4SLinus Torvalds dev_name, data_page); 28481da177e4SLinus Torvalds dput_out: 28492d92ab3cSAl Viro path_put(&path); 28501da177e4SLinus Torvalds return retval; 28511da177e4SLinus Torvalds } 28521da177e4SLinus Torvalds 2853537f7ccbSEric W. Biederman static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) 2854537f7ccbSEric W. Biederman { 2855537f7ccbSEric W. Biederman return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); 2856537f7ccbSEric W. Biederman } 2857537f7ccbSEric W. Biederman 2858537f7ccbSEric W. Biederman static void dec_mnt_namespaces(struct ucounts *ucounts) 2859537f7ccbSEric W. Biederman { 2860537f7ccbSEric W. Biederman dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); 2861537f7ccbSEric W. Biederman } 2862537f7ccbSEric W. Biederman 2863771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 2864771b1371SEric W. Biederman { 28656344c433SAl Viro ns_free_inum(&ns->ns); 2866537f7ccbSEric W. Biederman dec_mnt_namespaces(ns->ucounts); 2867771b1371SEric W. Biederman put_user_ns(ns->user_ns); 2868771b1371SEric W. Biederman kfree(ns); 2869771b1371SEric W. Biederman } 2870771b1371SEric W. Biederman 28718823c079SEric W. Biederman /* 28728823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 28738823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 28748823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 28758823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 28768823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 28778823c079SEric W. Biederman */ 28788823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 28798823c079SEric W. Biederman 2880771b1371SEric W. Biederman static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) 2881cf8d2c11STrond Myklebust { 2882cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 2883537f7ccbSEric W. Biederman struct ucounts *ucounts; 288498f842e6SEric W. Biederman int ret; 2885cf8d2c11STrond Myklebust 2886537f7ccbSEric W. Biederman ucounts = inc_mnt_namespaces(user_ns); 2887537f7ccbSEric W. Biederman if (!ucounts) 2888df75e774SEric W. Biederman return ERR_PTR(-ENOSPC); 2889537f7ccbSEric W. Biederman 2890cf8d2c11STrond Myklebust new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 2891537f7ccbSEric W. Biederman if (!new_ns) { 2892537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 2893cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 2894537f7ccbSEric W. Biederman } 28956344c433SAl Viro ret = ns_alloc_inum(&new_ns->ns); 289698f842e6SEric W. Biederman if (ret) { 289798f842e6SEric W. Biederman kfree(new_ns); 2898537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 289998f842e6SEric W. Biederman return ERR_PTR(ret); 290098f842e6SEric W. Biederman } 290133c42940SAl Viro new_ns->ns.ops = &mntns_operations; 29028823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 2903cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 2904cf8d2c11STrond Myklebust new_ns->root = NULL; 2905cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 2906cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 2907cf8d2c11STrond Myklebust new_ns->event = 0; 2908771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 2909537f7ccbSEric W. Biederman new_ns->ucounts = ucounts; 2910d2921684SEric W. Biederman new_ns->mounts = 0; 2911d2921684SEric W. Biederman new_ns->pending_mounts = 0; 2912cf8d2c11STrond Myklebust return new_ns; 2913cf8d2c11STrond Myklebust } 2914cf8d2c11STrond Myklebust 29150766f788SEmese Revfy __latent_entropy 29169559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 29179559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 29181da177e4SLinus Torvalds { 29196b3286edSKirill Korotaev struct mnt_namespace *new_ns; 29207f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 2921315fc83eSAl Viro struct mount *p, *q; 29229559f689SAl Viro struct mount *old; 2923cb338d06SAl Viro struct mount *new; 29247a472ef4SEric W. Biederman int copy_flags; 29251da177e4SLinus Torvalds 29269559f689SAl Viro BUG_ON(!ns); 29279559f689SAl Viro 29289559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 29299559f689SAl Viro get_mnt_ns(ns); 29309559f689SAl Viro return ns; 29319559f689SAl Viro } 29329559f689SAl Viro 29339559f689SAl Viro old = ns->root; 29349559f689SAl Viro 2935771b1371SEric W. Biederman new_ns = alloc_mnt_ns(user_ns); 2936cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 2937cf8d2c11STrond Myklebust return new_ns; 29381da177e4SLinus Torvalds 293997216be0SAl Viro namespace_lock(); 29401da177e4SLinus Torvalds /* First pass: copy the tree topology */ 29414ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 29429559f689SAl Viro if (user_ns != ns->user_ns) 2943132c94e3SEric W. Biederman copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; 29447a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 2945be34d1a3SDavid Howells if (IS_ERR(new)) { 2946328e6d90SAl Viro namespace_unlock(); 2947771b1371SEric W. Biederman free_mnt_ns(new_ns); 2948be34d1a3SDavid Howells return ERR_CAST(new); 29491da177e4SLinus Torvalds } 2950be08d6d2SAl Viro new_ns->root = new; 29511a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 29521da177e4SLinus Torvalds 29531da177e4SLinus Torvalds /* 29541da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 29551da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 29561da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 29571da177e4SLinus Torvalds */ 2958909b0a88SAl Viro p = old; 2959cb338d06SAl Viro q = new; 29601da177e4SLinus Torvalds while (p) { 2961143c8c91SAl Viro q->mnt_ns = new_ns; 2962d2921684SEric W. Biederman new_ns->mounts++; 29639559f689SAl Viro if (new_fs) { 29649559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 29659559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 2966315fc83eSAl Viro rootmnt = &p->mnt; 29671da177e4SLinus Torvalds } 29689559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 29699559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 2970315fc83eSAl Viro pwdmnt = &p->mnt; 29711da177e4SLinus Torvalds } 29721da177e4SLinus Torvalds } 2973909b0a88SAl Viro p = next_mnt(p, old); 2974909b0a88SAl Viro q = next_mnt(q, new); 29754ce5d2b1SEric W. Biederman if (!q) 29764ce5d2b1SEric W. Biederman break; 29774ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 29784ce5d2b1SEric W. Biederman p = next_mnt(p, old); 29791da177e4SLinus Torvalds } 2980328e6d90SAl Viro namespace_unlock(); 29811da177e4SLinus Torvalds 29821da177e4SLinus Torvalds if (rootmnt) 2983f03c6599SAl Viro mntput(rootmnt); 29841da177e4SLinus Torvalds if (pwdmnt) 2985f03c6599SAl Viro mntput(pwdmnt); 29861da177e4SLinus Torvalds 2987741a2951SJANAK DESAI return new_ns; 2988741a2951SJANAK DESAI } 2989741a2951SJANAK DESAI 2990cf8d2c11STrond Myklebust /** 2991cf8d2c11STrond Myklebust * create_mnt_ns - creates a private namespace and adds a root filesystem 2992cf8d2c11STrond Myklebust * @mnt: pointer to the new root filesystem mountpoint 2993cf8d2c11STrond Myklebust */ 29941a4eeaf2SAl Viro static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) 2995cf8d2c11STrond Myklebust { 2996771b1371SEric W. Biederman struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); 2997cf8d2c11STrond Myklebust if (!IS_ERR(new_ns)) { 29981a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 29991a4eeaf2SAl Viro mnt->mnt_ns = new_ns; 3000be08d6d2SAl Viro new_ns->root = mnt; 3001d2921684SEric W. Biederman new_ns->mounts++; 3002b1983cd8SAl Viro list_add(&mnt->mnt_list, &new_ns->list); 3003c1334495SAl Viro } else { 30041a4eeaf2SAl Viro mntput(m); 3005cf8d2c11STrond Myklebust } 3006cf8d2c11STrond Myklebust return new_ns; 3007cf8d2c11STrond Myklebust } 3008cf8d2c11STrond Myklebust 3009ea441d11SAl Viro struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) 3010ea441d11SAl Viro { 3011ea441d11SAl Viro struct mnt_namespace *ns; 3012d31da0f0SAl Viro struct super_block *s; 3013ea441d11SAl Viro struct path path; 3014ea441d11SAl Viro int err; 3015ea441d11SAl Viro 3016ea441d11SAl Viro ns = create_mnt_ns(mnt); 3017ea441d11SAl Viro if (IS_ERR(ns)) 3018ea441d11SAl Viro return ERR_CAST(ns); 3019ea441d11SAl Viro 3020ea441d11SAl Viro err = vfs_path_lookup(mnt->mnt_root, mnt, 3021ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 3022ea441d11SAl Viro 3023ea441d11SAl Viro put_mnt_ns(ns); 3024ea441d11SAl Viro 3025ea441d11SAl Viro if (err) 3026ea441d11SAl Viro return ERR_PTR(err); 3027ea441d11SAl Viro 3028ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 3029d31da0f0SAl Viro s = path.mnt->mnt_sb; 3030d31da0f0SAl Viro atomic_inc(&s->s_active); 3031ea441d11SAl Viro mntput(path.mnt); 3032ea441d11SAl Viro /* lock the sucker */ 3033d31da0f0SAl Viro down_write(&s->s_umount); 3034ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 3035ea441d11SAl Viro return path.dentry; 3036ea441d11SAl Viro } 3037ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 3038ea441d11SAl Viro 3039312db1aaSDominik Brodowski int ksys_mount(char __user *dev_name, char __user *dir_name, char __user *type, 3040312db1aaSDominik Brodowski unsigned long flags, void __user *data) 30411da177e4SLinus Torvalds { 3042eca6f534SVegard Nossum int ret; 3043eca6f534SVegard Nossum char *kernel_type; 3044eca6f534SVegard Nossum char *kernel_dev; 3045b40ef869SAl Viro void *options; 30461da177e4SLinus Torvalds 3047b8850d1fSTim Gardner kernel_type = copy_mount_string(type); 3048b8850d1fSTim Gardner ret = PTR_ERR(kernel_type); 3049b8850d1fSTim Gardner if (IS_ERR(kernel_type)) 3050eca6f534SVegard Nossum goto out_type; 30511da177e4SLinus Torvalds 3052b8850d1fSTim Gardner kernel_dev = copy_mount_string(dev_name); 3053b8850d1fSTim Gardner ret = PTR_ERR(kernel_dev); 3054b8850d1fSTim Gardner if (IS_ERR(kernel_dev)) 3055eca6f534SVegard Nossum goto out_dev; 30561da177e4SLinus Torvalds 3057b40ef869SAl Viro options = copy_mount_options(data); 3058b40ef869SAl Viro ret = PTR_ERR(options); 3059b40ef869SAl Viro if (IS_ERR(options)) 3060eca6f534SVegard Nossum goto out_data; 30611da177e4SLinus Torvalds 3062b40ef869SAl Viro ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); 3063eca6f534SVegard Nossum 3064b40ef869SAl Viro kfree(options); 3065eca6f534SVegard Nossum out_data: 3066eca6f534SVegard Nossum kfree(kernel_dev); 3067eca6f534SVegard Nossum out_dev: 3068eca6f534SVegard Nossum kfree(kernel_type); 3069eca6f534SVegard Nossum out_type: 3070eca6f534SVegard Nossum return ret; 30711da177e4SLinus Torvalds } 30721da177e4SLinus Torvalds 3073312db1aaSDominik Brodowski SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 3074312db1aaSDominik Brodowski char __user *, type, unsigned long, flags, void __user *, data) 3075312db1aaSDominik Brodowski { 3076312db1aaSDominik Brodowski return ksys_mount(dev_name, dir_name, type, flags, data); 3077312db1aaSDominik Brodowski } 3078312db1aaSDominik Brodowski 30791da177e4SLinus Torvalds /* 3080afac7cbaSAl Viro * Return true if path is reachable from root 3081afac7cbaSAl Viro * 308248a066e7SAl Viro * namespace_sem or mount_lock is held 3083afac7cbaSAl Viro */ 3084643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 3085afac7cbaSAl Viro const struct path *root) 3086afac7cbaSAl Viro { 3087643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 3088a73324daSAl Viro dentry = mnt->mnt_mountpoint; 30890714a533SAl Viro mnt = mnt->mnt_parent; 3090afac7cbaSAl Viro } 3091643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 3092afac7cbaSAl Viro } 3093afac7cbaSAl Viro 3094640eb7e7SMickaël Salaün bool path_is_under(const struct path *path1, const struct path *path2) 3095afac7cbaSAl Viro { 309625ab4c9bSYaowei Bai bool res; 309748a066e7SAl Viro read_seqlock_excl(&mount_lock); 3098643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 309948a066e7SAl Viro read_sequnlock_excl(&mount_lock); 3100afac7cbaSAl Viro return res; 3101afac7cbaSAl Viro } 3102afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 3103afac7cbaSAl Viro 3104afac7cbaSAl Viro /* 31051da177e4SLinus Torvalds * pivot_root Semantics: 31061da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 31071da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 31081da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 31091da177e4SLinus Torvalds * 31101da177e4SLinus Torvalds * Restrictions: 31111da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 31121da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 31131da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 31141da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 31151da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 31161da177e4SLinus Torvalds * 31174a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 31184a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 31194a0d11faSNeil Brown * in this situation. 31204a0d11faSNeil Brown * 31211da177e4SLinus Torvalds * Notes: 31221da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 31231da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 31241da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 31251da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 31261da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 31271da177e4SLinus Torvalds * first. 31281da177e4SLinus Torvalds */ 31293480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 31303480b257SHeiko Carstens const char __user *, put_old) 31311da177e4SLinus Torvalds { 31322d8f3038SAl Viro struct path new, old, parent_path, root_parent, root; 313384d17192SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt; 313484d17192SAl Viro struct mountpoint *old_mp, *root_mp; 31351da177e4SLinus Torvalds int error; 31361da177e4SLinus Torvalds 31379b40bc90SAl Viro if (!may_mount()) 31381da177e4SLinus Torvalds return -EPERM; 31391da177e4SLinus Torvalds 31402d8f3038SAl Viro error = user_path_dir(new_root, &new); 31411da177e4SLinus Torvalds if (error) 31421da177e4SLinus Torvalds goto out0; 31431da177e4SLinus Torvalds 31442d8f3038SAl Viro error = user_path_dir(put_old, &old); 31451da177e4SLinus Torvalds if (error) 31461da177e4SLinus Torvalds goto out1; 31471da177e4SLinus Torvalds 31482d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 3149b12cea91SAl Viro if (error) 3150b12cea91SAl Viro goto out2; 31511da177e4SLinus Torvalds 3152f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 315384d17192SAl Viro old_mp = lock_mount(&old); 315484d17192SAl Viro error = PTR_ERR(old_mp); 315584d17192SAl Viro if (IS_ERR(old_mp)) 3156b12cea91SAl Viro goto out3; 3157b12cea91SAl Viro 31581da177e4SLinus Torvalds error = -EINVAL; 3159419148daSAl Viro new_mnt = real_mount(new.mnt); 3160419148daSAl Viro root_mnt = real_mount(root.mnt); 316184d17192SAl Viro old_mnt = real_mount(old.mnt); 316284d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 3163fc7be130SAl Viro IS_MNT_SHARED(new_mnt->mnt_parent) || 3164fc7be130SAl Viro IS_MNT_SHARED(root_mnt->mnt_parent)) 3165b12cea91SAl Viro goto out4; 3166143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 3167b12cea91SAl Viro goto out4; 31685ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 31695ff9d8a6SEric W. Biederman goto out4; 31701da177e4SLinus Torvalds error = -ENOENT; 3171f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 3172b12cea91SAl Viro goto out4; 31731da177e4SLinus Torvalds error = -EBUSY; 317484d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 3175b12cea91SAl Viro goto out4; /* loop, on the same file system */ 31761da177e4SLinus Torvalds error = -EINVAL; 31778c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 3178b12cea91SAl Viro goto out4; /* not a mountpoint */ 3179676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 3180b12cea91SAl Viro goto out4; /* not attached */ 318184d17192SAl Viro root_mp = root_mnt->mnt_mp; 31822d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 3183b12cea91SAl Viro goto out4; /* not a mountpoint */ 3184676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 3185b12cea91SAl Viro goto out4; /* not attached */ 31864ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 318784d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 3188b12cea91SAl Viro goto out4; 31890d082601SEric W. Biederman /* make certain new is below the root */ 31900d082601SEric W. Biederman if (!is_path_reachable(new_mnt, new.dentry, &root)) 31910d082601SEric W. Biederman goto out4; 319284d17192SAl Viro root_mp->m_count++; /* pin it so it won't go away */ 3193719ea2fbSAl Viro lock_mount_hash(); 3194419148daSAl Viro detach_mnt(new_mnt, &parent_path); 3195419148daSAl Viro detach_mnt(root_mnt, &root_parent); 31965ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 31975ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 31985ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 31995ff9d8a6SEric W. Biederman } 32004ac91378SJan Blunck /* mount old root on put_old */ 320184d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 32024ac91378SJan Blunck /* mount new_root on / */ 320384d17192SAl Viro attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); 32046b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 32054fed655cSEric W. Biederman /* A moved mount should not expire automatically */ 32064fed655cSEric W. Biederman list_del_init(&new_mnt->mnt_expire); 32073895dbf8SEric W. Biederman put_mountpoint(root_mp); 3208719ea2fbSAl Viro unlock_mount_hash(); 32092d8f3038SAl Viro chroot_fs_refs(&root, &new); 32101da177e4SLinus Torvalds error = 0; 3211b12cea91SAl Viro out4: 321284d17192SAl Viro unlock_mount(old_mp); 3213b12cea91SAl Viro if (!error) { 32141a390689SAl Viro path_put(&root_parent); 32151a390689SAl Viro path_put(&parent_path); 3216b12cea91SAl Viro } 3217b12cea91SAl Viro out3: 32188c3ee42eSAl Viro path_put(&root); 3219b12cea91SAl Viro out2: 32202d8f3038SAl Viro path_put(&old); 32211da177e4SLinus Torvalds out1: 32222d8f3038SAl Viro path_put(&new); 32231da177e4SLinus Torvalds out0: 32241da177e4SLinus Torvalds return error; 32251da177e4SLinus Torvalds } 32261da177e4SLinus Torvalds 32271da177e4SLinus Torvalds static void __init init_mount_tree(void) 32281da177e4SLinus Torvalds { 32291da177e4SLinus Torvalds struct vfsmount *mnt; 32306b3286edSKirill Korotaev struct mnt_namespace *ns; 3231ac748a09SJan Blunck struct path root; 32320c55cfc4SEric W. Biederman struct file_system_type *type; 32331da177e4SLinus Torvalds 32340c55cfc4SEric W. Biederman type = get_fs_type("rootfs"); 32350c55cfc4SEric W. Biederman if (!type) 32360c55cfc4SEric W. Biederman panic("Can't find rootfs type"); 32370c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, 0, "rootfs", NULL); 32380c55cfc4SEric W. Biederman put_filesystem(type); 32391da177e4SLinus Torvalds if (IS_ERR(mnt)) 32401da177e4SLinus Torvalds panic("Can't create rootfs"); 3241b3e19d92SNick Piggin 32423b22edc5STrond Myklebust ns = create_mnt_ns(mnt); 32433b22edc5STrond Myklebust if (IS_ERR(ns)) 32441da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 32451da177e4SLinus Torvalds 32466b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 32476b3286edSKirill Korotaev get_mnt_ns(ns); 32481da177e4SLinus Torvalds 3249be08d6d2SAl Viro root.mnt = mnt; 3250be08d6d2SAl Viro root.dentry = mnt->mnt_root; 3251da362b09SEric W. Biederman mnt->mnt_flags |= MNT_LOCKED; 3252ac748a09SJan Blunck 3253ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 3254ac748a09SJan Blunck set_fs_root(current->fs, &root); 32551da177e4SLinus Torvalds } 32561da177e4SLinus Torvalds 325774bf17cfSDenis Cheng void __init mnt_init(void) 32581da177e4SLinus Torvalds { 325915a67dd8SRandy Dunlap int err; 32601da177e4SLinus Torvalds 32617d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 326220c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 32631da177e4SLinus Torvalds 32640818bf27SAl Viro mount_hashtable = alloc_large_system_hash("Mount-cache", 326538129a13SAl Viro sizeof(struct hlist_head), 32660818bf27SAl Viro mhash_entries, 19, 32673d375d78SPavel Tatashin HASH_ZERO, 32680818bf27SAl Viro &m_hash_shift, &m_hash_mask, 0, 0); 32690818bf27SAl Viro mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", 32700818bf27SAl Viro sizeof(struct hlist_head), 32710818bf27SAl Viro mphash_entries, 19, 32723d375d78SPavel Tatashin HASH_ZERO, 32730818bf27SAl Viro &mp_hash_shift, &mp_hash_mask, 0, 0); 32741da177e4SLinus Torvalds 327584d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 32761da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 32771da177e4SLinus Torvalds 32784b93dc9bSTejun Heo kernfs_init(); 32794b93dc9bSTejun Heo 328015a67dd8SRandy Dunlap err = sysfs_init(); 328115a67dd8SRandy Dunlap if (err) 328215a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 32838e24eea7SHarvey Harrison __func__, err); 328400d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 328500d26666SGreg Kroah-Hartman if (!fs_kobj) 32868e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 32871da177e4SLinus Torvalds init_rootfs(); 32881da177e4SLinus Torvalds init_mount_tree(); 32891da177e4SLinus Torvalds } 32901da177e4SLinus Torvalds 3291616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 32921da177e4SLinus Torvalds { 3293d498b25aSAl Viro if (!atomic_dec_and_test(&ns->count)) 3294616511d0STrond Myklebust return; 32957b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 3296771b1371SEric W. Biederman free_mnt_ns(ns); 32971da177e4SLinus Torvalds } 32989d412a43SAl Viro 32999d412a43SAl Viro struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) 33009d412a43SAl Viro { 3301423e0ab0STim Chen struct vfsmount *mnt; 3302e462ec50SDavid Howells mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, data); 3303423e0ab0STim Chen if (!IS_ERR(mnt)) { 3304423e0ab0STim Chen /* 3305423e0ab0STim Chen * it is a longterm mount, don't release mnt until 3306423e0ab0STim Chen * we unmount before file sys is unregistered 3307423e0ab0STim Chen */ 3308f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 3309423e0ab0STim Chen } 3310423e0ab0STim Chen return mnt; 33119d412a43SAl Viro } 33129d412a43SAl Viro EXPORT_SYMBOL_GPL(kern_mount_data); 3313423e0ab0STim Chen 3314423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 3315423e0ab0STim Chen { 3316423e0ab0STim Chen /* release long term mount so mount point can be released */ 3317423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 3318f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 331948a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 3320423e0ab0STim Chen mntput(mnt); 3321423e0ab0STim Chen } 3322423e0ab0STim Chen } 3323423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 332402125a82SAl Viro 332502125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 332602125a82SAl Viro { 3327143c8c91SAl Viro return check_mnt(real_mount(mnt)); 332802125a82SAl Viro } 33298823c079SEric W. Biederman 33303151527eSEric W. Biederman bool current_chrooted(void) 33313151527eSEric W. Biederman { 33323151527eSEric W. Biederman /* Does the current process have a non-standard root */ 33333151527eSEric W. Biederman struct path ns_root; 33343151527eSEric W. Biederman struct path fs_root; 33353151527eSEric W. Biederman bool chrooted; 33363151527eSEric W. Biederman 33373151527eSEric W. Biederman /* Find the namespace root */ 33383151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 33393151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 33403151527eSEric W. Biederman path_get(&ns_root); 33413151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 33423151527eSEric W. Biederman ; 33433151527eSEric W. Biederman 33443151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 33453151527eSEric W. Biederman 33463151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 33473151527eSEric W. Biederman 33483151527eSEric W. Biederman path_put(&fs_root); 33493151527eSEric W. Biederman path_put(&ns_root); 33503151527eSEric W. Biederman 33513151527eSEric W. Biederman return chrooted; 33523151527eSEric W. Biederman } 33533151527eSEric W. Biederman 33548654df4eSEric W. Biederman static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new, 33558654df4eSEric W. Biederman int *new_mnt_flags) 335687a8ebd6SEric W. Biederman { 33578c6cf9ccSEric W. Biederman int new_flags = *new_mnt_flags; 335887a8ebd6SEric W. Biederman struct mount *mnt; 3359e51db735SEric W. Biederman bool visible = false; 336087a8ebd6SEric W. Biederman 336144bb4385SAl Viro down_read(&namespace_sem); 336287a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 3363e51db735SEric W. Biederman struct mount *child; 336477b1a97dSEric W. Biederman int mnt_flags; 336577b1a97dSEric W. Biederman 33668654df4eSEric W. Biederman if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type) 3367e51db735SEric W. Biederman continue; 3368e51db735SEric W. Biederman 33697e96c1b0SEric W. Biederman /* This mount is not fully visible if it's root directory 33707e96c1b0SEric W. Biederman * is not the root directory of the filesystem. 33717e96c1b0SEric W. Biederman */ 33727e96c1b0SEric W. Biederman if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) 33737e96c1b0SEric W. Biederman continue; 33747e96c1b0SEric W. Biederman 3375a1935c17SEric W. Biederman /* A local view of the mount flags */ 337677b1a97dSEric W. Biederman mnt_flags = mnt->mnt.mnt_flags; 337777b1a97dSEric W. Biederman 3378695e9df0SEric W. Biederman /* Don't miss readonly hidden in the superblock flags */ 3379bc98a42cSDavid Howells if (sb_rdonly(mnt->mnt.mnt_sb)) 3380695e9df0SEric W. Biederman mnt_flags |= MNT_LOCK_READONLY; 3381695e9df0SEric W. Biederman 33828c6cf9ccSEric W. Biederman /* Verify the mount flags are equal to or more permissive 33838c6cf9ccSEric W. Biederman * than the proposed new mount. 33848c6cf9ccSEric W. Biederman */ 338577b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_READONLY) && 33868c6cf9ccSEric W. Biederman !(new_flags & MNT_READONLY)) 33878c6cf9ccSEric W. Biederman continue; 338877b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_ATIME) && 338977b1a97dSEric W. Biederman ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) 33908c6cf9ccSEric W. Biederman continue; 33918c6cf9ccSEric W. Biederman 3392ceeb0e5dSEric W. Biederman /* This mount is not fully visible if there are any 3393ceeb0e5dSEric W. Biederman * locked child mounts that cover anything except for 3394ceeb0e5dSEric W. Biederman * empty directories. 3395e51db735SEric W. Biederman */ 3396e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 3397e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 3398ceeb0e5dSEric W. Biederman /* Only worry about locked mounts */ 3399d71ed6c9SEric W. Biederman if (!(child->mnt.mnt_flags & MNT_LOCKED)) 3400ceeb0e5dSEric W. Biederman continue; 34017236c85eSEric W. Biederman /* Is the directory permanetly empty? */ 34027236c85eSEric W. Biederman if (!is_empty_dir_inode(inode)) 3403e51db735SEric W. Biederman goto next; 340487a8ebd6SEric W. Biederman } 34058c6cf9ccSEric W. Biederman /* Preserve the locked attributes */ 340677b1a97dSEric W. Biederman *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ 34078c6cf9ccSEric W. Biederman MNT_LOCK_ATIME); 3408e51db735SEric W. Biederman visible = true; 3409e51db735SEric W. Biederman goto found; 3410e51db735SEric W. Biederman next: ; 341187a8ebd6SEric W. Biederman } 3412e51db735SEric W. Biederman found: 341344bb4385SAl Viro up_read(&namespace_sem); 3414e51db735SEric W. Biederman return visible; 341587a8ebd6SEric W. Biederman } 341687a8ebd6SEric W. Biederman 34178654df4eSEric W. Biederman static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags) 34188654df4eSEric W. Biederman { 3419a1935c17SEric W. Biederman const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; 34208654df4eSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 34218654df4eSEric W. Biederman unsigned long s_iflags; 34228654df4eSEric W. Biederman 34238654df4eSEric W. Biederman if (ns->user_ns == &init_user_ns) 34248654df4eSEric W. Biederman return false; 34258654df4eSEric W. Biederman 34268654df4eSEric W. Biederman /* Can this filesystem be too revealing? */ 34278654df4eSEric W. Biederman s_iflags = mnt->mnt_sb->s_iflags; 34288654df4eSEric W. Biederman if (!(s_iflags & SB_I_USERNS_VISIBLE)) 34298654df4eSEric W. Biederman return false; 34308654df4eSEric W. Biederman 3431a1935c17SEric W. Biederman if ((s_iflags & required_iflags) != required_iflags) { 3432a1935c17SEric W. Biederman WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", 3433a1935c17SEric W. Biederman required_iflags); 3434a1935c17SEric W. Biederman return true; 3435a1935c17SEric W. Biederman } 3436a1935c17SEric W. Biederman 34378654df4eSEric W. Biederman return !mnt_already_visible(ns, mnt, new_mnt_flags); 34388654df4eSEric W. Biederman } 34398654df4eSEric W. Biederman 3440380cf5baSAndy Lutomirski bool mnt_may_suid(struct vfsmount *mnt) 3441380cf5baSAndy Lutomirski { 3442380cf5baSAndy Lutomirski /* 3443380cf5baSAndy Lutomirski * Foreign mounts (accessed via fchdir or through /proc 3444380cf5baSAndy Lutomirski * symlinks) are always treated as if they are nosuid. This 3445380cf5baSAndy Lutomirski * prevents namespaces from trusting potentially unsafe 3446380cf5baSAndy Lutomirski * suid/sgid bits, file caps, or security labels that originate 3447380cf5baSAndy Lutomirski * in other namespaces. 3448380cf5baSAndy Lutomirski */ 3449380cf5baSAndy Lutomirski return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && 3450380cf5baSAndy Lutomirski current_in_userns(mnt->mnt_sb->s_user_ns); 3451380cf5baSAndy Lutomirski } 3452380cf5baSAndy Lutomirski 345364964528SAl Viro static struct ns_common *mntns_get(struct task_struct *task) 34548823c079SEric W. Biederman { 345558be2825SAl Viro struct ns_common *ns = NULL; 34568823c079SEric W. Biederman struct nsproxy *nsproxy; 34578823c079SEric W. Biederman 3458728dba3aSEric W. Biederman task_lock(task); 3459728dba3aSEric W. Biederman nsproxy = task->nsproxy; 34608823c079SEric W. Biederman if (nsproxy) { 346158be2825SAl Viro ns = &nsproxy->mnt_ns->ns; 346258be2825SAl Viro get_mnt_ns(to_mnt_ns(ns)); 34638823c079SEric W. Biederman } 3464728dba3aSEric W. Biederman task_unlock(task); 34658823c079SEric W. Biederman 34668823c079SEric W. Biederman return ns; 34678823c079SEric W. Biederman } 34688823c079SEric W. Biederman 346964964528SAl Viro static void mntns_put(struct ns_common *ns) 34708823c079SEric W. Biederman { 347158be2825SAl Viro put_mnt_ns(to_mnt_ns(ns)); 34728823c079SEric W. Biederman } 34738823c079SEric W. Biederman 347464964528SAl Viro static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) 34758823c079SEric W. Biederman { 34768823c079SEric W. Biederman struct fs_struct *fs = current->fs; 34774f757f3cSAl Viro struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; 34788823c079SEric W. Biederman struct path root; 34794f757f3cSAl Viro int err; 34808823c079SEric W. Biederman 34810c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 3482c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || 3483c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) 3484ae11e0f1SZhao Hongjiang return -EPERM; 34858823c079SEric W. Biederman 34868823c079SEric W. Biederman if (fs->users != 1) 34878823c079SEric W. Biederman return -EINVAL; 34888823c079SEric W. Biederman 34898823c079SEric W. Biederman get_mnt_ns(mnt_ns); 34904f757f3cSAl Viro old_mnt_ns = nsproxy->mnt_ns; 34918823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 34928823c079SEric W. Biederman 34938823c079SEric W. Biederman /* Find the root */ 34944f757f3cSAl Viro err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, 34954f757f3cSAl Viro "/", LOOKUP_DOWN, &root); 34964f757f3cSAl Viro if (err) { 34974f757f3cSAl Viro /* revert to old namespace */ 34984f757f3cSAl Viro nsproxy->mnt_ns = old_mnt_ns; 34994f757f3cSAl Viro put_mnt_ns(mnt_ns); 35004f757f3cSAl Viro return err; 35014f757f3cSAl Viro } 35028823c079SEric W. Biederman 35034068367cSAndrei Vagin put_mnt_ns(old_mnt_ns); 35044068367cSAndrei Vagin 35058823c079SEric W. Biederman /* Update the pwd and root */ 35068823c079SEric W. Biederman set_fs_pwd(fs, &root); 35078823c079SEric W. Biederman set_fs_root(fs, &root); 35088823c079SEric W. Biederman 35098823c079SEric W. Biederman path_put(&root); 35108823c079SEric W. Biederman return 0; 35118823c079SEric W. Biederman } 35128823c079SEric W. Biederman 3513bcac25a5SAndrey Vagin static struct user_namespace *mntns_owner(struct ns_common *ns) 3514bcac25a5SAndrey Vagin { 3515bcac25a5SAndrey Vagin return to_mnt_ns(ns)->user_ns; 3516bcac25a5SAndrey Vagin } 3517bcac25a5SAndrey Vagin 35188823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 35198823c079SEric W. Biederman .name = "mnt", 35208823c079SEric W. Biederman .type = CLONE_NEWNS, 35218823c079SEric W. Biederman .get = mntns_get, 35228823c079SEric W. Biederman .put = mntns_put, 35238823c079SEric W. Biederman .install = mntns_install, 3524bcac25a5SAndrey Vagin .owner = mntns_owner, 35258823c079SEric W. Biederman }; 3526